Computer Vision utility functions

A library of Computer Vision and OpenCV utility functions. It also includes basic image processing and common utility functions.

Two major contributions are:

1. Template matching
2. Image Utilities

Requirements:

• Numpy
• OpenCV
• matplotlib
• scipy, scikit-image

Template Matching utils

One drawback with OpenCV template matching module is that it is limited to RGB or similar 3-channel feature representations only. However, more meaningful representations like HOG or DeepNetwork features are represented with more number of channels.

Usage

The package is pip installable. The easiest way to install is by using pip.

$pip install cv_utils

You can also download the git repo and install by running the below command.

$python setup.py install

HOG template matching

An example for template matching with HOG features. It returns both the heatmap and the scale factor by which the feature representation reduced the image dimension.

Other supported features are:

1. HOG
2. Color spaces: gray, rgb, lab, luv, hsv, hls

import cv2 as cv
from cv_utils import template_matching as tm

template = cv.imread('tests/resources/kelloggs-red-fruit.jpg')
image = cv.imread('tests/resources/sch-image.jpg')

options = dict(feature='hog')
heatmap, scale = tm.feature_match(template, image, options)

Multi features template matching

In some cases, using one feature is not enough to represent the entire image. Combining RGB with HOG might perform better in some cases.

import ...

template = cv.imread('tests/resources/kelloggs-red-fruit.jpg')
image = cv.imread('tests/resources/sch-image.jpg')

options = dict()
options['features'] = [dict(feature='hog'), dict(feature='rgb')]
heatmap, scale = tm.multi_feat_match(template, image, options)

Normal template matching to find single instance

In most cases, template matching is used to find exactly one instance of an object in the search image. A special utility for that is given below. It returns the bounding box object and the chosen minimum heatmap score.

import ...

template = cv.imread('tests/resources/kelloggs-red-fruit.jpg')
image = cv.imread('tests/resources/sch-image.jpg')

box, score = tm.match_one(template, image, dict(feature='rgb'))

Image Utilities

Remove background

img = cv.imread('tests/resources/with-bg.jpg')
img_res = img_utils.remove_bg(img)

Add background with padding

img = cv.imread('tests/resources/without-black-bg.jpg')
img_res = img_utils.add_bg(img, 50, cv_utils.COL_YELLOW)

Please look at the img_utils file to know more about all the utility functions. Not all the features are covered in this doc.

Other utility functions

• collage (Constructs a collage of same-sized images with specified padding)
• imshow (matplotlib plotting for multiple images)
• is_gray (Is gray-scale image)
• add_rect (Add bounding box in an image)
• add_text_img (Add text in an image)