diff --git a/.gitignore b/.gitignore index b15c6ef..8155012 100644 --- a/.gitignore +++ b/.gitignore @@ -1,3 +1,3 @@ *.jpg *.npz - +__pycache__/ diff --git a/.vscode/launch.json b/.vscode/launch.json index 54a210c..9f2f4c1 100644 --- a/.vscode/launch.json +++ b/.vscode/launch.json @@ -10,11 +10,6 @@ "request": "launch", "program": "${file}", "console": "integratedTerminal", - "args": [ - "./SmTIAS*_white01.JPG", - "--roi", "190", "920", "1440", "1440", - "--sigma", "10.0", - ] } ] } \ No newline at end of file diff --git a/calc_illum_dist.py b/calc_illum_dist.py index 8e77cd1..5e7e052 100644 --- a/calc_illum_dist.py +++ b/calc_illum_dist.py @@ -5,55 +5,37 @@ import cv2 import numpy as np -# # Load the color image -# names = ["SmTIAS001", "SmTIAS002", "SmTIAS003"] -# imgs = np.array([cv2.imread(f"{name}_white01.jpg") for name in names]) -# names.append("mean") - -# for i in range(len(names)): -# if i < len(names) - 1: -# img = imgs[i] -# else: -# img = np.mean(imgs, axis=0).astype("uint8") -# # print("img shape", img.shape) - -sigma_default = 10.0 +import config -def calc_illum_dist(args, img, imgname): +def calc_illum_dist(img, roi_pos): # Convert the image to grayscale (luminance) gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # ROI生成 - roi_x, roi_y, roi_w, roi_h = 0, 0, gray.shape[1], gray.shape[0] - if args.roi: - roi_x = args.roi[0] - roi_y = args.roi[1] - roi_w = args.roi[2] - roi_h = args.roi[3] - gray_roi = gray[roi_y : roi_y + roi_h, roi_x : roi_x + roi_w] - vmin = gray_roi.min() - vmax = gray_roi.max() - vmean = gray_roi.mean() - print(f"{imgname} value min,mean,max", vmin, vmean, vmax) + roi_x1, roi_y1, roi_x2, roi_y2 = ( + roi_pos[0][0], + roi_pos[0][1], + roi_pos[1][0], + roi_pos[1][1], + ) + gray_roi = gray[roi_y1:roi_y2, roi_x1:roi_x2] + + # ROI描画 + anno = img.copy() + cv2.rectangle(anno, (roi_x1, roi_y1), (roi_x2, roi_y2), (0, 0, 255), 2) # 相対照度分布の算出 + vmean = gray_roi.mean() ratio_roi = gray_roi / vmean - print( - f"{imgname} ratio min,mean,max", - ratio_roi.min(), - ratio_roi.mean(), - ratio_roi.max(), - ) - ratio = np.full(gray.shape, -1.0, dtype=np.float32) - ratio[roi_y : roi_y + roi_h, roi_x : roi_x + roi_w] = ratio_roi - - # Normalize the grayscale image - normalized = 255.0 * (gray_roi - vmin) / (vmax - vmin) - normalized = cv2.GaussianBlur( - normalized, ksize=(0, 0), sigmaX=args.sigma, sigmaY=args.sigma - ) - normalized = normalized.astype("uint8") + # print( + # "ratio min,mean,max", + # ratio_roi.min(), + # ratio_roi.mean(), + # ratio_roi.max(), + # ) + # ratio = np.full(gray.shape, -1.0, dtype=np.float32) + # ratio[roi_y1:roi_y2, roi_x1:roi_x2] = ratio_roi # if i == len(names) - 1: # @@ -67,48 +49,133 @@ # # print("ratio_whole max", ratio_whole.max()) # np.savez_compressed("illum_dist.npz", ratio_whole) + return ratio_roi, anno + + +def detect_landmarks(img): + # ランドマーク検出 + landmarks = [] + annotation = img.copy() + for i in range(len(config.corner_imgs_file)): + template = cv2.imread(config.corner_imgs_file[i]) + roi_w = int(config.corner_search_size[0] * img.shape[1]) + roi_h = int(config.corner_search_size[1] * img.shape[0]) + roi_x = int(img.shape[1] - roi_w) if i % 2 == 1 else 0 + roi_y = int(config.corner_search_pos[i] * img.shape[0]) + roi = img[roi_y : roi_y + roi_h, roi_x : roi_x + roi_w] + res = cv2.matchTemplate(roi, template, cv2.TM_CCOEFF_NORMED) + _, _, _, max_loc = cv2.minMaxLoc(res) + cv2.rectangle( + annotation, + (roi_x, roi_y), + (roi_x + roi_w, roi_y + roi_h), + (0, 255, 0), + 2, + ) + cv2.rectangle( + annotation, + (max_loc[0] + roi_x, max_loc[1] + roi_y), + ( + max_loc[0] + roi_x + template.shape[1], + max_loc[1] + roi_y + template.shape[0], + ), + (0, 0, 255), + 2, + ) + landmarks.append( + ( + max_loc[0] + roi_x + (template.shape[1] if i < 2 else 0), + max_loc[1] + roi_y + template.shape[0] // 2, + ) + ) + # print(max_loc) + # disp = cv2.resize(img, None, fx=0.5, fy=0.5) + # cv2.imshow("img", disp) + return landmarks, annotation + + +def registration(img, target, src): + # ホモグラフィ変換 + homograpy, _ = cv2.findHomography(np.array(src), np.array(target), cv2.RANSAC, 5.0) + # print(homograpy) + return cv2.warpPerspective(img, homograpy, (img.shape[1], img.shape[0])) + + +def resize_show(name, img): + fx = config.display_size[0] / img.shape[1] + fy = config.display_size[1] / img.shape[0] + scale = min(fx, fy) + scale = 1.0 if scale > 1.0 else scale + disp = cv2.resize(img, None, fx=scale, fy=scale) + cv2.imshow(name, disp) + + +def show_heatmap(name, dist): + # Normalize the grayscale image + vmin = dist.min() + vmax = dist.max() + normalized = 255.0 * (dist - vmin) / (vmax - vmin) + normalized = cv2.GaussianBlur( + normalized, ksize=(0, 0), sigmaX=config.sigma, sigmaY=config.sigma + ) + normalized = normalized.astype("uint8") + # Create a heatmap from the grayscale image heatmap = cv2.applyColorMap(normalized, cv2.COLORMAP_JET) - heatmap = cv2.resize(heatmap, None, fx=0.5, fy=0.5) - - cv2.imshow(f"{imgname} heatmap", heatmap) - - -def parse_arguments(): - parser = argparse.ArgumentParser(description="Calculate illumination distribution") - parser.add_argument("input_files", nargs="+", help="Input image files") - parser.add_argument( - "--roi", type=int, nargs=4, help="region of interest (x, y, w, h)" - ) - parser.add_argument( - "--sigma", - type=float, - # nargs=1, - help=f"sigma of gaussian blur (default: {sigma_default})", - default=sigma_default, - ) - return parser.parse_args() + resize_show(name, heatmap) + # heatmap = cv2.resize(heatmap, None, fx=0.5, fy=0.5) def main(): + # メイン関数 + # 実行時引数からデータ生成 - args = parse_arguments() - file_names = [f for file in args.input_files for f in glob.glob(file)] + # args = parse_arguments() + file_names = [f for file in config.input_files for f in glob.glob(file)] names = [Path(f).stem for f in file_names] imgs = [cv2.imread(file) for file in file_names] - # 平均画像の生成 - if len(imgs) > 1: - img_mean = np.mean(np.array(imgs), axis=0).astype("uint8") - imgs.append(img_mean) - names.append("mean") - + # 位置合わせランドマーク検出 + landmarks = [] + annotations = [] for i in range(len(imgs)): - calc_illum_dist(args, imgs[i], names[i]) + lm, anno = detect_landmarks(imgs[i]) + landmarks.append(lm) + annotations.append(anno) + + if len(imgs) > 1: + # 位置合わせ + for i in range(1, len(imgs)): + imgs[i] = registration(imgs[i], landmarks[0], landmarks[i]) + annotations[i] = registration(annotations[i], landmarks[0], landmarks[i]) + + # 平均画像の生成 + img_mean = np.mean(np.array(imgs), axis=0).astype("uint8") + imgs.append(img_mean) + names.append("mean") + # resize_show("mean", img_mean) + + # ランドマークアノテーションの重ね合わせ + anno_mean = np.mean(np.array(annotations), axis=0).astype("uint8") + resize_show("mean annotation", anno_mean) + + # 照度分布算出 + roi_pos = [landmarks[0][0], landmarks[0][3]] + for i in range(0, len(imgs)): + # print(names[i]) + ratio_roi, anno = calc_illum_dist(imgs[i], roi_pos) + show_heatmap(f"{names[i]} heatmap", ratio_roi) + resize_show(f"{names[i]} anno_roi", anno) + if names[i] == "mean": + vmin, vmax, _, _ = cv2.minMaxLoc(ratio_roi) + print(f"ratio min,max={vmin:.3f},{vmax:.3f}") + np.savez_compressed(config.output_file, ratio_roi) + print(f"'{config.output_file}' saved.") cv2.waitKey(0) cv2.destroyAllWindows() if __name__ == "__main__": + # 起動 main() diff --git a/config.py b/config.py new file mode 100644 index 0000000..d514d20 --- /dev/null +++ b/config.py @@ -0,0 +1,19 @@ +# 入力画像 +input_files = ["imgs/SmTIAS*_white01.JPG"] +# 出力比率画像 +output_file = "illum_dist.npz" +# template_file = "imgs/SmTIAS_tmpl.JPG" +# ランドマーク画像 +corner_imgs_file = [ + "imgs/corner_ul.jpg", + "imgs/corner_ur.jpg", + "imgs/corner_bl.jpg", + "imgs/corner_br.jpg", +] +# ランドマーク探索領域 +corner_search_pos = [650 / 3264, 630 / 3264, 2100 / 3264, 2100 / 3264] +corner_search_size = [500 / 1836, 600 / 3264] +# 輝度分布のスムージング係数 +sigma = 10.0 +# 表示最大 幅，高さ +display_size = [800, 1200]