Imagerie

Lightweight python package grouping together useful computer vision functions and operations

This package was initially created for myself to extract some of the most useful important functions,
from packages like scipy, skimage, etc...

This was done in order to expose all these very useful computer vision functions,
and be able to use them in a Serverless environment.

Documentation

Prerequisites

This package only supports python3 today.

Only tested on python3.6 and python3.7

Installation

Install using pip pip install imagerie.

Functions

imagerie.get_rotation(pt1, pt2)

Returns the rotation value in degrees between 2 (x,y) points.

imagerie.order_points(points)

You've guessed it, this function simply sorts a list of 4 [x, y] coordinates in a clockwise manner, starting from the top-left.

imagerie.remove_lonely_small_objects(grayscale)

This function removes small white objects from a binary mask.

imagerie.remove_smaller_objects(grayscale)

Finds all contours and removes them all except the biggest one.

imagerie.biggest_contour(grayscale)

Finds and retrieves the biggest contour from a grayscale image.

imagerie.get_biggest_contour(contours)

Simply retrieves and returns the biggest contour from a given list of contours.

imagerie.closest_point(point: tuple, points)

Returns the closest (x, y) point from a given list of (x, y) points/coordinates.

imagerie_lite.midpoint(ptA, ptB)

Calculates the X,Y middle points from provided 2 points.

imagerie_lite.line_intersection(line1: tuple, line2: tuple)

Returns the intersection point between two lines.

imagerie.get_corners(grayscale, middle_points=False, centroid=False, max_corners=4, quality_level=0.01, min_distance=15)

Returns the (x, y) coordinates of the 4 corners of a rectangular shaped object from binary mask by default. However, you can also calculate the top and bottom middle coordinates by providing middle_points=True. And by providint centroid=True, you can get the (x, y) coordinates of the center.

imagerie.warp_perspective(image, src_pts: list, dst_pts: list)

Performs a cv2.warpPerspective() operation and expects 2 lists of (x, y) corner points of the source and destination image.

imagerie.warp_homography(image, src_pts: list, dst_pts: list, method=cv2.RANSAC, reproj_threshold=5.0)

Performs a cv2.warpPerspective() operation after cv2.findHomography().

imagerie.image_composite_with_mask(to_add: PIL.Image.Image, destination: PIL.Image.Image, mask: PIL.Image.Image)

Combines the to_add and destination images, to_add image will be added on top of destination image and only the white area from the mask image will be retained from to_add image.

imagerie.combine_two_images_with_mask(background_img, foreground_img, mask)

Combines the images with the help of the provided mask. Note that only the white area of the mask will be selected from the foreground_img.

imagerie.prepare_for_prediction_single(img, shape=(768, 768), as_array=True)

Loads and resizes a single image to a given shape (default: 768, 768) and returns it by default as a numpy array.

imagerie.prepare_for_prediction(imgs, shape=(768, 768))

Does the same thing as imagerie.prepare_for_prediction_single but for multiple images.

imagerie.fill_holes(gray_img: ndarray, min=200, max=255)

Fills black pixel holes that reside inside of a binary object.

min represents the thresh used in cv2.threshold(gray, thresh=200, ...).

max represents the maxval used in cv2.threshold(gray, maxval=255, ...).

imagerie.img_as_uint(img)

Converts image to 16 bit unsigned integer format.

imagerie.img_as_float(img)

Convert an image to floating point format.

imagerie.normalize_binary_img(img: np.ndarray)

Very useful when re-converting a predicted binary mask img to original grayscale format.

Credits

• Ibragim Abubakarov <ibragim.ai95@gmail.com>