preprocess image for ocr

preprocess_image_ocr.py

import cv2
import numpy as np
import PIL.Image as Image
import imutils


def increase_brightness(image, value=10):
    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
    h, s, v = cv2.split(hsv)

    lim = 255 - value
    v[v > lim] = 255
    v[v <= lim] += value

    final_hsv = cv2.merge((h, s, v))
    output = cv2.cvtColor(final_hsv, cv2.COLOR_HSV2BGR)

    return output


def perspective_transform(image, hull):
    height, width = image.shape[:2]
    image_corner = hull.reshape(4, 2)
    image_corner = np.float32(image_corner)
    center_x, center_y = np.mean(image_corner, axis=0)
    sort_key = np.select(
        [
            np.logical_and(
                image_corner[:, 0] < center_x, image_corner[:, 1] < center_y
            ),
            np.logical_and(
                image_corner[:, 0] > center_x, image_corner[:, 1] < center_y
            ),
            np.logical_and(
                image_corner[:, 0] > center_x, image_corner[:, 1] > center_y
            ),
            np.logical_and(
                image_corner[:, 0] < center_x, image_corner[:, 1] > center_y
            ),
        ],
        [0.0, 1.0, 2.0, 3.0],
    )
    sort_key = np.expand_dims(sort_key, axis=0)
    temp = np.concatenate((image_corner, sort_key.T), axis=1)
    sorted_corner = temp[np.lexsort(temp.T)]
    sorted_corner = sorted_corner[:, 0:2]

    dst_corner = np.array(
        ((0, 0), (width - 1, 0), (width - 1, height - 1), (0, height - 1)),
        np.float32,
    )

    M, _ = cv2.findHomography(sorted_corner, dst_corner, cv2.RANSAC, 10)
    warped = cv2.warpPerspective(image, M, (width, height))

    return warped


def alignment(image, margin=30):
    thresh = cv2.threshold(image[:, :, 3], 0, 255, cv2.THRESH_BINARY)[1]
    blur = cv2.medianBlur(thresh, 15)
    cnts = cv2.findContours(blur, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    cnts = imutils.grab_contours(cnts)
    cnts = sorted(cnts, key=cv2.contourArea, reverse=True)
    perimeter = cv2.arcLength(cnts[0], True)
    approx = cv2.approxPolyDP(cnts[0], 0.02 * perimeter, True)
    hull = cv2.convexHull(approx, clockwise=True)
    warped = perspective_transform(image, hull)
    height, width = warped.shape[:2]
    output = warped[margin : height - margin, margin : width - margin]

    return output


def remove_shadow(image):
    image = increase_brightness(image)
    rgb_planes = cv2.split(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
    result_norm_planes = []
    for plane in rgb_planes:
        dilated_img = cv2.dilate(plane, np.ones((5, 5), np.uint8))
        opening_img = cv2.morphologyEx(dilated_img, cv2.MORPH_OPEN, np.ones((11, 11)))
        closing_img = cv2.morphologyEx(opening_img, cv2.MORPH_CLOSE, np.ones((11, 11)))
        diff_img = 255 - cv2.absdiff(plane, closing_img)
        norm_img = cv2.normalize(
            diff_img,
            None,
            alpha=0,
            beta=255,
            norm_type=cv2.NORM_MINMAX,
            dtype=cv2.CV_8UC1,
        )
        result_norm_planes.append(norm_img)

    result_norm = cv2.merge(result_norm_planes)

    return result_norm