本文针对OCR cls模型中,不同尺寸图片resize后得到的图像可能会进行不均匀拉伸,导致图像不清晰。因此考虑在图像resize前先对图像进行padding操作,使之实现均匀缩放。(本人正在探索阶段,该想法未必可以提高准确率)
上代码:
import os
import cv2
import numpy as np
import argparse
def process(input_dir, out_dir):
for filename in os.listdir(input_dir):
file, suffix = os.path.splitext(filename)
if suffix.lower() not in [".jpg", ".jpeg", ".png"]:
continue
old_path = os.path.join(input_dir, filename)
img = cv2.imread(old_path)
img_H, img_W, img_C = img.shape
ratio = 192 // 48
if img_H * ratio > img_W:
im_padding = np.zeros((img_H, img_H*ratio, img_C), dtype=np.float32)
im_padding[:, 0:img_W, :] = img
elif img_H * ratio == img_W:
im_padding = img
else:
im_padding = img[:, 0:img_H*ratio, :]
new_path = os.path.join(out_dir, filename)
resize_img = cv2.resize(im_padding, (192, 48))
print(resize_img.shape)
cv2.imwrite(new_path, resize_img)
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument('--input_dir', default='../paddleocr/PaddleOCR/train_data/cls/test/')
parser.add_argument('--out_dir', default='../paddleocr/PaddleOCR/train_data/cls/test_t/')
return parser.parse_args()
def main(args):
if not os.path.isdir(args.out_dir):
os.makedirs(args.out_dir)
process(args.input_dir, args.out_dir)
if __name__ == "__main__":
main(get_args())