utils.py

import math
import cv2
import matplotlib.pyplot as plt
import numpy as np
import numpy.ma as ma


# Distance to rectangle
def distance(rect, point):
    dx = max(min(rect[0][0], rect[1][0]) - point[0], 0, point[0] - max(rect[0][0], rect[1][0]))
    dy = max(min(rect[0][1], rect[1][1]) - point[1], 0, point[1] - max(rect[0][1], rect[1][1]))
    # dx = max(rect.bottom_left.x - point[0], 0, point[0] - (rect.bottom_left.x+rect.width))
    # dy = max(rect.bottom_left.y - point[1], 0, point[1] - (rect.bottom_left.y+rect.height))
    return math.sqrt(dx*dx + dy*dy)


# Resize image
def imresize(img):
    if img.shape[1] < img.shape[0]: # portrait or landscape?
        coef = 300/img.shape[1]
    else:
        coef = 300/img.shape[0]
    # if coef < 1: return img # make sure not to make it bigger

    dim = (int(img.shape[1]*coef), int(img.shape[0]*coef))
    return cv2.resize(img, dim, interpolation=cv2.INTER_AREA), 1/coef


# Highlight image
def highlight(img, pt1, pt2):
    img = img.copy()

    alpha = 2.2 # Simple contrast control
    beta = 30    # Simple brightness control

    hl = img[pt1[1]:pt2[1],pt1[0]:pt2[0]]
    hl = np.clip(alpha*hl + beta, 0, 255) # option 1: numpy
    # hl = cv2.convertScaleAbs(hl, alpha=alpha, beta=beta) # option 2: cv2
    img[pt1[1]:pt2[1],pt1[0]:pt2[0]] = hl
    print('dtype of img:', img.dtype, '', 'dtype of hl:', hl.dtype)

    return img
# modified from https://docs.opencv.org/3.4/d3/dc1/tutorial_basic_linear_transform.html


# add mask for selective image manipulation
def add_mask(mask, pt1, pt2):
    mask[pt1[1]:pt2[1],pt1[0]:pt2[0]] = 1
    return mask


# add mask for selective image manipulation
def apply_highlight(img, weak_mask, strong_mask):
    # weak -= np.logical_and(weak, strong) # remove overlap # unnecessary because strong overwrites weak

    alpha = 2.4 # Simple contrast control
    beta = 30    # Simple brightness control

    weak = np.clip(img * alpha/1.7 + beta/2, 0, 255).astype(np.uint8)
    strong = np.clip(img * alpha + beta, 0, 255).astype(np.uint8)

    highlighted = np.where(weak_mask, weak, img)
    highlighted = np.where(strong_mask, strong, highlighted)

    return highlighted


def main():
    # distance
    rectangle = [(-1,1),(1,-1)]
    point_1 = (0,0) # should be true
    point_2 = (1,1) # should be true ideally (outline inclusive)
    point_3 = (1.1,1.1) # should be false

    print(distance(rectangle, point_1))
    print(distance(rectangle, point_2))
    print(distance(rectangle, point_3))

    # masking
    a = np.zeros(shape=[4,4])
    a = add_mask(a, (2,2), (3,3))
    print(a)

    folder = './demo_image/'
    filename = 'demodemo2.jpg'
    img = cv2.imread(folder+filename)
    
    # highlighted = highlight(img, (926, 1478), (2021, 1756))
    # print(highlighted.dtype)
    # plt.imshow(highlighted)
    weak = add_mask(np.zeros(shape=img.shape, dtype=np.uint8), (926, 1478), (2021, 1756))
    strong = add_mask(np.zeros(shape=img.shape, dtype=np.uint8), (1026, 1578), (2121, 1856))
    plt.imshow(apply_highlight(img, weak, strong)) # test
    plt.show()


if __name__ == "__main__": main()