imagewizard

imagewizard

Source hosted at github: https://github.com/PriceSpider-NeuIntel/imagewizard/

imagewizard is a python based library for performing various image manipulations and operations,

1. Image Hashing <https://github.com/PriceSpider-NeuIntel/imagewizard#image-hashing>_

   • Average Hashing <https://github.com/PriceSpider-NeuIntel/imagewizard#average-hash-a-hash>_
   • Distance Hashing <https://github.com/PriceSpider-NeuIntel/imagewizard#distance-hash-d-hash>_
   • Perception Hashing <https://github.com/PriceSpider-NeuIntel/imagewizard#perception-hash-p-hash>_
   • Wavelet Hashing <https://github.com/PriceSpider-NeuIntel/imagewizard#wavelet-hash-w-hash>_

2. Image Similarity (hash distance computation) <https://github.com/PriceSpider-NeuIntel/imagewizard#image-similarity-hash-distance>_

   • Hamming Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#hamming-distance>_
   • Cosine Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#cosine-distance>_
   • Euclidean Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#euclidean-distance>_
   • Manhattan Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#manhattan-distance>_
   • Jaccard Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#jaccard-distance>_
   • Minkowski Distance <https://github.com/PriceSpider-NeuIntel/imagewizard#minkowski-distance>_

3. Image Processing and Transformations <https://github.com/PriceSpider-NeuIntel/imagewizard#image-processing--transformations>_

   • Segmentation <https://github.com/PriceSpider-NeuIntel/imagewizard#segmentation>_
   • Resize/scale <https://github.com/PriceSpider-NeuIntel/imagewizard#resize>_
   • Convert to gray scale <https://github.com/PriceSpider-NeuIntel/imagewizard#gray-scale>_
   • Rotate <https://github.com/PriceSpider-NeuIntel/imagewizard#rotate>_
   • Crop <https://github.com/PriceSpider-NeuIntel/imagewizard#crop>_
   • Mirror <https://github.com/PriceSpider-NeuIntel/imagewizard#mirror>_
   • Blur <https://github.com/PriceSpider-NeuIntel/imagewizard#blur>_
   • Luminosity (Brightness) <https://github.com/PriceSpider-NeuIntel/imagewizard#luminosity>_
   • Skew <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---perspective>_
     • Perspective <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---perspective>_
     • Affine <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---affine>_

4. Image Analysis <https://github.com/PriceSpider-NeuIntel/imagewizard#image-analysis>_

   • Dominant colors <https://github.com/PriceSpider-NeuIntel/imagewizard#dominant-colors>_
   • Average/Mean Color <https://github.com/PriceSpider-NeuIntel/imagewizard#averagemean-color>_
   • Frequent/Mode Color <https://github.com/PriceSpider-NeuIntel/imagewizard#frequent-color>_
   • Trim/Crop to content <https://github.com/PriceSpider-NeuIntel/imagewizard#trimcrop-to-content>_

Image Hashing

Intuition

---

Given an imput image, imagewizard can compute a hash for the image based on it visual appearance. It is understood that images that are perceptually similar must have similar hashes as well. The similarity here is a metric that we can choose to compute on, generally hamming distance is considered, but we can choose other distance metrics too. By utilizing imagewizard we can find near-identical images in constant time, or at worst, O(log n).

imagewizard supports the following hashing techniques:

• average hashing (a Hash_)
• perception hashing (p Hash_)
• difference hashing (d Hash_)
• wavelet hashing (w Hash_)

Basic Usage

---

Let's take a look at how we can implement image hashing using imagewizard. We will use PIL/Pillow and OpenCv2-python image libraries to read an image and get a 8x8 hash.

Step 1: read an image file

---

::

import cv2 from PIL import Image pil_image = Image.open('test.png') cv2_image = cv2.imread('test.png')

Step 2: perform image hashing

---

Remember, if you are using opencv2 for reading an image file, the order of the color channels are BGR while that for a PIL Image its RGB. The channel information has to be passed as a parameter to the function while performing hashing. The default value of order is RGB

imagewizard.Hashing()

• .ahash(image, hash_size, order)
• .dhash(image, hash_size, order) * hamming distance between 0 - 10, would indicate the images being compared are similar
• .phash(image, hash_size, order)
• .whash(image, hash_size, order)

Parameters:

• image - must be a PIL instance image or numpy array in RGB or opencv image in BGR
• hash_size - (integer) default 8 for 64 bit hash
• order - (string) RGB, BGR: defaults to 'RGB' - input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import imagewizard as iw iw_hash = iw.Hashing()

Average hash (a hash)

---

a_hash_pil = iw_hash.ahash(image = pil_image, hash_size = 8, order = 'RGB') a_hash_cv2 = iw_hash.ahash(image = cv2_image, hash_size = 8, order = 'BGR')

print("PIL a-hash: {}".format(a_hash_pil)) PIL a-hash: fefff80000000000 print("cv2 a-hash: {}".format(a_hash_cv2)) cv2 a-hash: fefff80000000000

Distance hash (d hash)

---

• hamming distance between 0 - 10, would indicate the images being compared are similar

d_hash_pil = iw_hash.dhash(image = pil_image, hash_size = 8, order = 'RGB') d_hash_cv2 = iw_hash.dhash(image = cv2_image, hash_size = 8, order = 'BGR')

print("PIL d-hash: {}".format(a_hash_pil)) PIL d-hash: 48b09035b16c9ccb print("cv2 d-hash: {}".format(a_hash_cv2)) cv2 d-hash: 48b09035b16c9ccb

Perception hash (p hash)

---

p_hash_pil = iw_hash.phash(image = pil_image, hash_size = 8, order = 'RGB') p_hash_cv2 = iw_hash.phash(image = cv2_image, hash_size = 8, order = 'BGR')

print("PIL p-hash: {}".format(p_hash_pil)) PIL p-hash: d0ddd594473657c0 print("cv2 p-hash: {}".format(p_hash_cv2)) cv2 p-hash: d0ddd594473657c0

Wavelet hash (w hash)

---

w_hash_pil = iw_hash.whash(image = pil_image, hash_size = 8, order = 'RGB') w_hash_cv2 = iw_hash.whash(image = cv2_image, hash_size = 8, order = 'BGR')

print("PIL w-hash: {}".format(w_hash_pil)) PIL w-hash: fffffe90100e4420 print("cv2 w-hash: {}".format(w_hash_cv2)) cv2 w-hash: fffffe90100e4420

Few other operations

---

To get the hash value, simply cast the returned object to str,

hash_value1 = str(a_hash_cv2) hash_value2 = str(a_hash_pil)

You can also find the hamming distance (the number of bit positions in which the two bits are different) by simply applying subtraction operation,

hash_diff = a_hash_pil - a_hash_pil print(hash_diff) 0

Since the two hashes are of the same image, the hamming distance is 0. For more information on hamming distance - https://en.wikipedia.org/wiki/Hamming_distance

If you simply want to check if the two hashes are exact matches, you could do that too,

print(a_hash_pil == a_hash_cv2) True print(a_hash_cv2 == d_hash_cv2) False

Image Similarity (hash distance)

Now that we have a hash corresponsding to an image, we can find how similar other images are, by comparing the hashes, i.e, finding the hash distances. Lower the values, more similar are the images. imagewizard provides various distance algorithms for computing hash distances between two hashes,

imagewizard.Similarity().similarity(hash1, hash2, metric = )

The value can be one of the following-

• hamming
• cosine
• euclidean
• manhattan
• jaccard
• minkowski

Basic Usage

---

import imagewizard as iw import cv2 iw_hash = iw.Hashing() iw_similarity = iw.Similarity()

image1 = cv2.imread('test.png') hash1_str = str(iw_hash.dhash(image1, order = 'BGR')) image2 = cv2.imread('test2.png') hash2_str = str(iw_hash.dhash(image2, order = 'BGR'))

Hamming distance

---

print("hamming: ", iw_similarity.similarity(hash1_str, hash2_str, metric = 'hamming')) hamming: 26

Cosine distance

---

print("cosine: ", iw_similarity.similarity(hash1_str, hash2_str, metric = 'cosine')) cosine: 0.546

Euclidean distance

---

print("euclidean : {}".format(iw_similarity.similarity(hash1_str, hash2_str, metric = 'euclidean'))) euclidean : 5.0

Manhattan distance

---

print("manhattan : {}".format(iw_similarity.similarity(hash1_str, hash2_str, metric = 'manhattan'))) manhattan : 26

Jaccard distance

---

print("jaccard : {}".format(iw_similarity.similarity(hash1_str, hash2_str, metric = 'jaccard'))) jaccard : 1.0

Minkowski distance

---

p value is set to 3 while computing minkowski distance

print("minkowski : {}".format(iw_similarity.similarity(hash1_str, hash2_str, metric = 'minkowski'))) minkowski : 2.924

Concise explanation of distance algorithms_

Image Processing & Transformations

imagewizard provides the following image processing and transformations

• Segmentation <https://github.com/PriceSpider-NeuIntel/imagewizard#segmentation>_
• Resize/scale <https://github.com/PriceSpider-NeuIntel/imagewizard#resize>_
• Convert to gray scale <https://github.com/PriceSpider-NeuIntel/imagewizard#gray-scale>_
• Rotate <https://github.com/PriceSpider-NeuIntel/imagewizard#rotate>_
• Crop <https://github.com/PriceSpider-NeuIntel/imagewizard#crop>_
• Mirror <https://github.com/PriceSpider-NeuIntel/imagewizard#mirror>_
• Blur <https://github.com/PriceSpider-NeuIntel/imagewizard#blur>_
• Luminosity (Brightness) <https://github.com/PriceSpider-NeuIntel/imagewizard#luminosity>_
• Skew <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---perspective>_ * Perspective <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---perspective>_ * Affine <https://github.com/PriceSpider-NeuIntel/imagewizard#skew---affine>_

Segmentation

---

imagewizard provides methods for image segmentation, i.e, reconstructing a given image with a given set of colors alone. Every pixel in the original image is mapped to its nearest color from the set of colors and reconstructed. The following code demonstrates image segmentation of the famous picture of lenna with three colors (RGB values),

• [224 166 147]
• [110 34 71]
• [195 98 100]

imagewizard.Processing().segmentation(img, rgb_list: [[int]], order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• rgb_list: 2 dimensional np array with shape (n,3) 3 being the channel values in order RGB, eg: [[224, 166, 147], [110, 34, 71], [195, 98, 100]]
• order: (RGB, BGR) input order of the colors BGR/RGB. Deafult order: RGB Note: The output will be a numpy.array of the same order

cv_img = cv.imread('data/original_images/lenna.png') pil_img = Image.open('data/original_images/lenna.png') ip = imagewizard.Processing() rgb_colors_list = [[224, 166, 147], [110, 34, 71], [195, 98, 100]]

================ ================ ================ Color 1 Color 2 Color 3 ================ ================ ================ |cv_dom_c0| |cv_dom_c1| |cv_dom_c2|
================ ================ ================

cv_result = ip.segmentation(cv_img, rgb_colors_list, 'bgr') pil_result = ip.segmentation(pil_img, rgb_colors_list, 'rgb') pil_res_im = Image.fromarray(pil_res)

cv2.imshow("original image", cv_img) cv2.imshow('Segmented Image', cv_result) pil_res_im.show()

=============== =============== Original Segmented Image =============== =============== |lenna_org| |segmented_im| =============== ===============

Resize

---

imagewizard provides methods to resize/scale an image to desired pixel (width x height),

imagewizard.Processing().resize(img, interpolation_method: str, resize_width: int, resize_height: int, resize_percentage: float, order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• interpolation_method: (s, z) s/shrink or z/zoom; default to shrink
• resize_percentage: (0, 100) floating value. to resize image by the specified percentage
• resize_width, resize_height: (in pixels) if unspecified, defaults to 50% of original img width & height. If either only width or height is specified, the other dimension is scaled implicitly, to keep the aspect ratio intact.
  Note: these will be ignored if resize_percentage is specified
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'
  Note: The output will be a numpy.array of the same order

Lets put resize to work on an image of the beautiful view outside Mumbai T2

======== ====================================== Original 50% of original - Aspect Ratio Intact ======== ====================================== |t2_img| |t2_r3|
======== ======================================

================ ==================================== 300px by 300px height: 200px - Aspect Ratio Intact ================ ==================================== |t2_r1| |t2_r2|
================ ====================================

.. |t2_img| image:: tests/data/original_images/street.png :width: 450

Resize Image to 50% height X width, keeping aspect ratio intact

img = cv2.imread('data/test.png') ip = imagewizard.Processing()
res = ip.resize(img, resize_percentage = 50, order = 'bgr') cv2.imshow('Resized Image', res)

.. |t2_r3| image:: tests/data/processed_images/resize/shrink-50-percent.png :width: 60%

Resize Image to 300px by 300px

img = cv2.imread('data/test.png') ip = imagewizard.Processing()
res = ip.resize(img, resize_width=300, resize_height=300, order = 'bgr') cv2.imshow('Resized Image', res)

.. |t2_r1| image:: tests/data/processed_images/resize/shrink-300px-300px.png :width: 100px :height: 100px

Resize Image to height 200px, keeping aspect ratio intact

img = cv2.imread('data/test.png') ip = imagewizard.Processing()
res = ip.resize(img, resize_height=200, order = 'bgr') cv2.imshow('Resized Image', res)

.. |t2_r2| image:: tests/data/processed_images/resize/shrink-200px.png :width: 60%

Gray scale

---

imagewizard provides methods to convert a given color image to gray scale/inverted in various forms such as,

• Inverted Colors
• To Gray/Gray Inverted
• To Binary/Binary Inverted
• To Zero/Zero Inverted
• To Truncated/Truncated Inverted

imagewizard.Processing().img2grayscale(image, to_binary: bool, to_zero: bool, inverted: bool, trunc: bool, is_gray: bool, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)

• thresholding_options * to_binary: (True/False) - defaults to False, converts the image to a complete black and white image without any shade of gray * to_zero: (True/False) - defaults to False, converts an image to zero thresholding if set to True * trunc: (True/False) - defaults to False, converts an image to truncated thresholding if set to True * inverted: (True/False) - defaults to False, this parameter can be used along with any of the above parameter. If set to True, the colorspace will be inverted * is_gray: (True/False) - defaults to True, if set to false and used along with ('inverted' == True) the colorspace of the image will be inverted

  Note: the preference of the parameters follows - truc > to_binary > to_zero. The lower order parameter will be ignored in presence of a parameter with a greater preference. 
  

• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'
  Note: The output will be a numpy.array of the same order

Let us use the famous picture of Lena, to demonstrate gray scaling.

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

inverted_img = ip.img2grayscale(img, inverted=True, is_gray=False, order = 'bgr') cv.imshow("inverted Image", inverted_img)

================ ================ Original Inverted ================ ================ |lenna_org| |clr_inv|
================ ================

gray_image = ip.img2grayscale(img, order = 'bgr') cv2.imshow("Gray", gray_image)

gray_inv_image = ip.img2grayscale(img, inverted=True, order = 'bgr') cv.imshow("Gray Inverted", gray_inv_image)

================ ================ Gray Gray Inv
================ ================ |gray| |gray_inv| ================ ================

trunc_image = ip.img2grayscale(img, trunc=True, order = 'bgr') cv.imshow("Trucated Threshold", trunc_image)

trunc_inv_image = ip.img2grayscale(img, trunc=True, inverted=True, order = 'bgr') cv.imshow("Trucated Threshold Inv", trunc_inv_image)

================ ================ Truncated Truncated Inv ================ ================ |trunc| |trunc_inv| ================ ================

binary_image = ip.img2grayscale(img, to_binary=True, order = 'bgr') cv2.imshow("Binary Threshold", binary_image)

binary_inv_image = ip.img2grayscale(img, to_binary=True, inverted=True, order = 'bgr') cv2.imshow("Binary Threshold Inverted", binary_inv_image)

================ ================
Binary Binary Inv
================ ================ |bin_img| |bin_inv| ================ ================

to_zero_image = ip.img2grayscale(img, to_zero=True, order = 'bgr') cv2.imshow("To Zero", to_zero_image)

to_zero_inverted = ip.img2grayscale(img, to_zero=True, inverted = True, order = 'bgr') cv2.imshow("To Zero Inverted", to_zero_inverted)

================ ================ To Zero To Zero Inv ================ ================ |tz| |tz_inv| ================ ================

Rotate

---

imagewizard provides method to rotate a given image, with or without scaling. The image provided is rotated in anti-clockwise direction by the rotation angle in degree specified.

• ip.Processing().rotate(image, rotation_degree: float, scaling_factor: float, order: str)

Parameters:

• image: (numpy.array, PIL.image, cv2.image)
• rotation_degree: rotation angle (in degrees), the image will be rotate in anti-clockwise direction
• scaling_factor: scale the image to desired factor. set to 1.0 to maintain the original scale of the image. 0.5 to halve the size of the image, to double the size of the image, use 2.0.
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

Following code demonstrates rotation,

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

rotate_by_90 = ip.rotate(img, rotation_degree = 90, order='bgr') cv2.imshow("Rotate by 90 degrees", rotate_by_90)

rotate_by_180 = ip.rotate(img, rotation_degree = 180, order='bgr') cv2.imshow("Rotate by 180 degrees", rotate_by_180)

rotate_by_270 = ip.rotate(img, rotation_degree = 270, order='bgr') cv2.imshow("Rotate by 270 degrees", rotate_by_270)

rotate_by_315_scale = ip.rotate(img, rotation_degree = 315, scaling_factor=0.5, order='bgr') cv2.imshow("Rotate by 315 degrees, scale 0.5x", rotate_by_315_scale)

rotate_by_45_scale = ip.rotate(img, rotation_degree = 45, scaling_factor=2, order='bgr') cv2.imshow("Rotate by 45 degrees, scale 2x", rotate_by_45_scale)

================ ================ ================ Original 90 deg 180 deg
================ ================ ================ |lenna_org| |90deg| |180deg|
================ ================ ================

================ ================= =================== 270 deg 45 deg, scale 2x 315 deg, scale 0.5x
================ ================= =================== |270deg| |45degs| |315degs|
================ ================= ===================

Crop

---

imagewizard lets you crop a given image. Provide the starting and ending, X and Y coordinates to crop the image to.

imagewizard.Processing().crop(img: Image, start_x: float, end_x: float, start_y: float, end_y: float, is_percentage: Bool, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• start_x: starting pixel coordinate along the x-axis/width of the image
• end_x: ending pixel coordinate along the x-axis/width of the image
• start_y: starting pixle coordinate along the y-axis/height of the image
• end_y: ending pixle coordinate along the y-axis/height of the image
• is_percentage: if True, the coordinates will be considered as percentages, default: False
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

crop1 = ip.crop(img, start_x = 50, end_x = 100, start_y = 50, end_y = 100, is_percentage = True, order='bgr') cv2.imshow("Crop % (a)", crop1)

crop2 = ip.crop(img, start_x = 400, end_x = 1000, start_y = 0, end_y = 500, is_percentage = False, order='bgr') cv2.imshow("Crop by px", crop2)

crop3 = ip.crop(img, start_x = 0, end_x = 50, start_y = 0, end_y = 50, is_percentage = True, order='bgr') cv2.imshow("Crop % (b)", crop3)

================ ================= ================= =================== Original Crop % (a) Crop by px Crop % (b)
================ ================= ================= =================== |t2_img| |crop1| |crop2| |crop3|
================ ================= ================= ===================

Mirror

---

imagewizard provides methods to mirror/flip a given image. The image can be flipped around its X-axis or Y-axis or both X and Y axis by providing the flip_code parameter. The following code demonstrates flipping around various axes.

imagewizard.Processing().mirror(img: Image, flip_code: int, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• flip_code:
  • = 0 for flipping the image around the y-axis (vertical flipping);

  • 0 for flipping around the x-axis (horizontal flipping);

  • < 0 for flipping around both axes
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

mir_x = ip.mirror(img, flip_code=1, order='bgr') cv.imshow('Horizontal Mirror (X)', mir_x)

mir_y = ip.mirror(img, flip_code=0, order='bgr') cv.imshow('Vertical Mirror (Y)', mir_y)

mir_xy = ip.mirror(img, flip_code=-1, order='bgr') cv.imshow('Mirrored both X and Y', mir_xy)

======================== ======================== ======================== ======================== Original Horizontal Mirror (X) Vertical Mirror (Y) Mirrored both X and Y ======================== ======================== ======================== ======================== |lenna_org| |mir_x| |mir_y| |mir_xy|
======================== ======================== ======================== ========================

Blur

---

imagewizard provides methods to blur a given image. The intensity of the blur can be passed as an argument to the function. The following code demonstrates blurring.

imagewizard.Processing().blur(img: Image, blur_level: int, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• blur_level: (int, > 0 and < 100,000) intensity of blur
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

blur_5 = ip.blur(img, blur_level = 5, order='bgr') cv.imshow('Blur level 5', blur_5)

blur_25 = ip.blur(img, blur_level = 25, order='bgr') cv.imshow('Blur level 25', blur_25)

blur_50 = ip.blur(img, blur_level = 50, order='bgr') cv.imshow('Blur level 50', blur_50)

============= ============= ============= ============= Original Blur level 5 Blur level 25 Blur level 50 ============= ============= ============= ============= |t2_img| |blur_5| |blur_25| |blur_50|
============= ============= ============= =============

Luminosity

---

imagewizard provides methods to change the luminosity/brightness of a given image. The intensity of the brightness can be passed as an argument to the function. A positive intensity value will brighten the image, whereas a negative value will darken the image. The following code demonstrates changing the brightness levels.

imagewizard.Processing().luminosity(img: Image, intensity_shift: int, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• intensity_shift: -ve value to darken and +ve value to brighten
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

lum_100 = ip.luminosity(img, intensity_shift = 100, order = 'bgr') cv.imshow('Brightness level increased by 100', lum_100)

lum_neg_100 = ip.luminosity(img, intensity_shift = -100, order = 'bgr') cv.imshow('Brightness level decreased by 100', lum_neg_100)

================================= ================================= ================================= Brightness level decreased by 100 Original Brightness level increased by 100 ================================= ================================= ================================= |lum_neg_100| |lenna_org| |lum_100|
================================= ================================= =================================

Skew - Perspective

---

imagewizard provides methods to perspective tranform an image. You need to provide 4 points on the input image and corresponding points on the output image. Among these 4 points, 3 of them should not be collinear. Following code demonstrates Perspective Transformation.

imagewizard.Processing().skew_perspective(img: Image, input_points: np.float32, output_points: np.float32, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• input_points: four points on input image, ex: np.float32([[x1,y1],[x2,y2],[x3,y3],[x4,y4]]), (xi, yi are floating point)
• output_points: four points on output location correspoinding to input_points' to be transformed, ex: np.float32([[p1,q1],[p2,q2],[p3,q3],[p4,q4]]), (pi, qi are floating point)
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

input_points = np.float32([(100, 320), (472, 156), (250, 580), (630, 345)]) output_points = np.float32([[0,0], [500,0], [0,350], [500,350]])

skew_img = ip.skew_perspective(img, input_points = input_points, output_points = output_points, order = 'bgr') cv.imshow('Perspective Transformation', skew_img)

================================= ================================= Original Perspective Transformation
================================= ================================= |skew_per_org| |skew_per_tf|
================================= =================================

• The green points on the input image specifies the coordinates of the pixels that will be mapped to output points.
• The coordinates passed in the code above are in the order - TOP LEFT, TOP RIGHT, BOTTOM LEFT, BOTTOM RIGHT
• The corresponding input pixel coordinates are - TL:(100, 320), TR:(472, 156), BL:(250, 580), BR:(630, 345)]
• The corresponding output pixel coordinates are - TL:(0, 0), TR:(500, 0), BL:(0, 350), BR:(500, 350)]

Skew - Affine

---

imagewizard provides methods to affine transform an image. In affine transformation, all parallel lines in the original image will still be parallel in the output image. Provide three points from input image and their corresponding locations in output image. Following code demonstrates Affine Transformation.

imagewizard.Processing().affine(img: Image, input_points: np.float32, output_points: np.float32, order: str)

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• input_points: three points on input image, ex: np.float32([[x1,y1],[x2,y2],[x3,y3]]), (xi, yi are floating point)
• output_points: three points on output location correspoinding to input_points' to be transformed, np.float32([[p1,q1],[p2,q2],[p3,q3]]), (pi, qi are floating point)
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 img = cv2.imread('original_image.png') ip = imagewizard.Processing()

input_points = np.float32([[50,50],[200,50],[50,200]]) output_points = np.float32([[10,100],[200,50],[100,250]])

skew_img = ip.skew_perspective(img, input_points = input_points, output_points = output_points, order = 'bgr') cv.imshow('Affine Transformation', skew_img)

================================= ================================= Original Affine Transformation
================================= ================================= |skew_aff_org| |skew_aff_tf|
================================= =================================

• The green points on the input image specifies the coordinates of the pixels that will be mapped to output points.
• The coordinates passed in the code above are in the order - TOP LEFT, TOP RIGHT, BOTTOM LEFT
• The corresponding input pixel coordinates are - TL:(50, 50), TR:(200, 50), BL:(50, 200)]
• The corresponding output pixel coordinates are - TL:(10, 100), TR:(200, 50), BL:(100, 250)]

For more information check this documentation <https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_geometric_transformations/py_geometric_transformations.html#affine-transformation>_

Image Analysis

Dominant Colors

---

imagewizard provides methods to find the 'n' dominant colors in an image. Following code demonstrates using the function to get 'n' dominant colors in an image.

imagewizard.Analysis().dominant_colors(img: Image, no_of_colors: int = 3, order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• no_of_colors: (int) no of dominant colors RGB to return
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

import cv2 from PIL import Image img_cv = cv2.imread('original_image.png') img_pil = Image.open("original_image.png")

imanalysis = imagewizard.Analysis() img_cv_result = imanalysis.dominant_colors(img_pil, 3, 'bgr') img_pil_result = imanalysis.dominant_colors(img_cv, 3, 'rgb')

print("CV image - dominant colors (RGB) : ", img_cv_result) CV image - dominant colors (RGB) : [[224 166 147] [110 34 71] [195 98 100]]

============= ================ ================ ================ ================ Original Clustered Image Color 1 Color 2 Color 3 ============= ================ ================ ================ ================ |lenna_org| |clustered_im| |cv_dom_c0| |cv_dom_c1| |cv_dom_c2|
============= ================ ================ ================ ================

• The Clustered Image can be constructed with Segmentation <https://github.com/PriceSpider-NeuIntel/imagewizard#segmentation>_ method of imagewizard

print("PIL image - dominant colors (RGB) : ", img_pil_result) PIL image - dominant colors (RGB) : [[224 166 147] [110 34 71] [195 98 100]]

============= ================ ================ ================ ================ Original Clustered Image Color 1 Color 2 Color 3 ============= ================ ================ ================ ================ |lenna_org| |clustered_im| |pil_dom_c0| |pil_dom_c1| |pil_dom_c2|
============= ================ ================ ================ ================

Average/Mean Color

---

imagewizard provides methods that calculates and returns the mean/average color of an image Following code demonstrates using the function to get the average pixel color in RGB

imagewizard.Analysis().mean_color(img: Image, order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

Returns:

• Tuple of RGB values of the mean color calculated

import cv2 from PIL import Image img_cv = cv2.imread('original_image.png') img_pil = Image.open("original_image.png")

imanalysis = imagewizard.Analysis() img_cv_result = imanalysis.mean_color(img_pil, 'bgr') img_pil_result = imanalysis.mean_color(img_cv, 'rgb')

print("PIL image - mean color (RGB) :", img_cv_result) PIL image - mean color (RGB) : (180, 99, 105)

print("CV image - mean color (RGB) :", img_cv_result) CV image - mean color (RGB) : (180, 99, 105)

=================== =================== Original Average/Mean Color
=================== ===================
|lenna_org_ave| |lenna_result_ave| =================== ===================

Frequent Color

---

imagewizard provides methods that calculates and returns the frequent/mode color of an image Following is the demonstration code,

imagewizard.Analysis().frequent_color(img: Image, order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

Returns:

• Tuple of RGB values of the mode color calculated

import cv2 from PIL import Image img_cv = cv2.imread('original_image.png') img_pil = Image.open("original_image.png")

imanalysis = imagewizard.Analysis() img_cv_result = imanalysis.frequent_color(img_pil, 'bgr') img_pil_result = imanalysis.frequent_color(img_cv, 'rgb')

print("PIL image - frequent color (RGB) :", img_cv_result) PIL image - frequent color (RGB) : (88, 18, 60)

print("CV image - frequent color (RGB) :", img_cv_result) CV image - frequent color (RGB) : (88, 18, 60)

=================== ===================
Original Frequent/Mode Color
=================== ===================
|lenna_org_mode| |lenna_result_mode| =================== ===================

Trim/Crop to Content

---

imagewizard provides methods to Trim/Crop an image to its content (removes uniform color spaced padding around the image) Following code demonstrates using the function to trim an image

imagewizard.Analysis().trim_to_content(img: Image, order: str = 'rgb')

Parameters:

• img: (numpy.array, PIL.image, cv2.image)
• order: (RGB, BGR) input order of the colors. If using PIL to read an image, 'order' need not be specified. If opencv is used to read an image, 'order' must be set to 'BGR'

Returns:

• PIL/numpy.array of the order specified

import cv2 from PIL import Image img_cv = cv2.imread('original_image.png') img_pil = Image.open("original_image.png")

imanalysis = imagewizard.Analysis() img_cv_result = imanalysis.trim_to_content(img_cv, 'bgr') img_pil_result = imanalysis.trim_to_content(img_pil)

cv.imshow("original", img_cv) cv.imshow("Trimmed Image", img_cv_result) img_pil_result.show()

================ ================= Original Trimmed Image

---

|quite_flow_org| |quite_flow_trim| ================ =================

================ ================= Original Trimmed Image

---

|san_disk_org| |san_disk_trim| ================ =================

.. _a Hash: http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html .. _p Hash: http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html .. _d Hash: http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html .. _w Hash: https://fullstackml.com/2016/07/02/wavelet-image-hash-in-python/ .. _distance algorithms: https://dataconomy.com/2015/04/implementing-the-five-most-popular-similarity-measures-in-python/ .. _pypi: https://pypi.python.org/pypi/

.. |lenna_org| image:: tests/data/original_images/lenna.png

.. |clr_inv| image:: tests/data/processed_images/gray/clr_inverted.png

.. |gray| image:: tests/data/processed_images/gray/gray.png

.. |gray_inv| image:: tests/data/processed_images/gray/gray_inverted.png

.. |bin_img| image:: tests/data/processed_images/gray/binary_img.png

.. |bin_inv| image:: tests/data/processed_images/gray/binary_inv_img.png

.. |tz| image:: tests/data/processed_images/gray/to_zero_img.png

.. |tz_inv| image:: tests/data/processed_images/gray/to_zero_inv.png

.. |trunc| image:: tests/data/processed_images/gray/trunc_img.png

.. |trunc_inv| image:: tests/data/processed_images/gray/trunc_inverted.png

.. |90deg| image:: tests/data/processed_images/rotate/rotate-90deg.png

.. |180deg| image:: tests/data/processed_images/rotate/rotate-180deg.png

.. |270deg| image:: tests/data/processed_images/rotate/rotate-270deg.png

.. |315degs| image:: tests/data/processed_images/rotate/rotate-315deg-scale.png

.. |45degs| image:: tests/data/processed_images/rotate/rotate-45deg-scale.png

.. |crop1| image:: tests/data/processed_images/crop/crop1.png

.. |crop2| image:: tests/data/processed_images/crop/crop2.png

.. |crop3| image:: tests/data/processed_images/crop/crop3.png

.. |mir_x| image:: tests/data/processed_images/mirror/flip_x.png

.. |mir_y| image:: tests/data/processed_images/mirror/flip_y.png

.. |mir_xy| image:: tests/data/processed_images/mirror/flip_xy.png

.. |blur_5| image:: tests/data/processed_images/blur/blur5.png

.. |blur_25| image:: tests/data/processed_images/blur/blur25.png

.. |blur_50| image:: tests/data/processed_images/blur/blur50.png

.. |lum_100| image:: tests/data/processed_images/luminosity/lum_100.png

.. |lum_neg_100| image:: tests/data/processed_images/luminosity/lum_neg_100.png

.. |skew_per_org| image:: tests/data/original_images/skew_per_org.png

.. |skew_per_tf| image:: tests/data/processed_images/skew/skew_per.png

.. |skew_aff_org| image:: tests/data/original_images/skew_aff_org.png

.. |skew_aff_tf| image:: tests/data/processed_images/skew/skew_aff.png

.. ########## dominant colors ########### .. |pil_dom_c0| image:: tests/data/analysed_images/dominant/centroid_color_0.jpg .. |pil_dom_c1| image:: tests/data/analysed_images/dominant/centroid_color_1.jpg .. |pil_dom_c2| image:: tests/data/analysed_images/dominant/centroid_color_2.jpg .. |cv_dom_c0| image:: tests/data/analysed_images/dominant/centroid_color_0_cv.jpg .. |cv_dom_c1| image:: tests/data/analysed_images/dominant/centroid_color_1_cv.jpg .. |cv_dom_c2| image:: tests/data/analysed_images/dominant/centroid_color_2_cv.jpg .. |clustered_im| image:: tests/data/analysed_images/dominant/clustered_image.jpg

.. ########## mean colors ########### .. |lenna_org_ave| image:: tests/data/analysed_images/mean/lenna-original-ave.jpg .. |lenna_result_ave| image:: tests/data/analysed_images/mean/lenna-result-ave.jpg

.. ########## frerquent colors ########### .. |lenna_org_mode| image:: tests/data/analysed_images/mode/lenna-original-mode.jpg .. |lenna_result_mode| image:: tests/data/analysed_images/mode/lenna-result-mode.jpg

.. ########## crop/trim to content ########### .. |quite_flow_trim| image:: tests/data/analysed_images/crop_to_content/trimmed_quite_flow10.png .. |san_disk_trim| image:: tests/data/analysed_images/crop_to_content/trimmed_san_disk.png

.. |quite_flow_org| image:: tests/data/original_images/quiet_flow10.png .. |san_disk_org| image:: tests/data/original_images/san_disk_white_pad.png

.. |segmented_im| image:: tests/data/processed_images/segmented_image.png