Character segmentation for dataset preparation

character_segmentation.py

"""
-- Ashok Kumar Pant ([email protected])
-- Treeleaf Technologies Pvt. Ltd.
-- Date: 1/18/20
"""
import argparse
import os
from random import random
from traceback import print_exc

import cv2


def _get_boxes(img):
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    (thresh, img_bin) = cv2.threshold(gray, 240, 255, cv2.THRESH_BINARY)
    img_bin = 255 - img_bin
    v_ker = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
    dilated = cv2.dilate(img_bin, v_ker)
    contours, hierarchy = cv2.findContours(dilated, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
    table_boxes = []
    for c in contours:
        x, y, w, h = cv2.boundingRect(c)
        bbox = [x, y, x + w, y + h]
        if w * h >= 0.5 * (img.shape[0] * img.shape[1]):
            continue
        if w > 100 and h > 100:  # Minimum area check
            table_boxes.append(bbox)
    return table_boxes


def _save_images(img, boxes, output_dir, prefix=""):
    for i, bbox in enumerate(boxes):
        l, t, r, b = bbox
        chip = img[t:b, l:r]
        cv2.imwrite(os.path.join(output_dir, prefix + "_" + str(i) + ".jpg"), chip)


def segment_characters(filename, output_dir, prefix=""):
    try:
        img = cv2.imread(filename)
        boxes = _get_boxes(img)
        os.makedirs(output_dir, exist_ok=True)
        _save_images(img, boxes, output_dir, prefix=prefix)
        return True
    except Exception as _:
        print_exc()


def _list_images(directory, shuffle=False):
    """
    Generic function to list images from directory or file
    :return:  list of images in a given directory or file
    """
    _ext = ['.jpg', '.jpeg', '.bmp', '.png', '.JPG', '.JPEG', '.ppm', '.pgm', '.webp']
    _images = []
    for subdir, dirs, files in os.walk(directory):
        for file in files:
            file_path = os.path.join(subdir, file)
            if file_path.endswith(tuple(_ext)):
                _images.append(file_path)
    if shuffle:
        random.shuffle(_images)
    return _images


def _is_dir(path):
    return os.path.isdir(path)


if __name__ == '__main__':
    parser = argparse.ArgumentParser(description="Character segmentation")
    parser.add_argument('--input', help="Input image file or directory", default=None)
    parser.add_argument('--output', help="Output directory", default="tmp/")
    parser.add_argument('--shuffle', help="Shuffle images", default=False)
    args = parser.parse_args()
    if args.input is None:
        print("Input image file or directory to process!")
        print("Example: python character_segmentation.py --input image.jpg --output tmp/")
        https://github.com/ashokpant/ai_sapphire/blob/master/images/image.jpg
        parser.print_help()
        exit(0)

    if _is_dir(args.input):
        images = _list_images(args.input, args.shuffle)
    else:
        images = [args.input]
    for i, filename in enumerate(images):
        success = segment_characters(filename, args.output, prefix=str(i + 1))
        if not success:
            print("Unable to process the file: {}".format(filename))
        if i > 0 and i % 10 == 0:
            print("Processed {} files: ".format(i))
    print("Processed total {} files".format(len(images)))
    print("Done.")