计算机视觉作业基于直方图的自适应阈值分割、聚类实现纹理图像分割、模板匹配、目标跟踪、背景建模、目标检测python源码.zip

# coding: utf-8 # Author：WangTianRui # Date ：2020/12/14 9:20 import matplotlib.pyplot as plt import matplotlib.image as mpimg import numpy as np import os, cv2, time from mpl_toolkits.mplot3d import Axes3D def load_img(path): if os.path.exists(path): return np.array(mpimg.imread(path), dtype="float32").astype(np.uint8) else: print("图片路径错误") return def plot_3d(x, y, z, title): fig = plt.figure() ax = Axes3D(fig) x, y = np.meshgrid(x, y) ax.plot_surface(x, y, z, rstride=1, cstride=1, cmap='rainbow') # zdir 表示向那个轴投影 # ax.contourf(x, y, img, zdir='z', offset=np.max(z) + 0.5, cmap='gray') ax.set_title(title) plt.show() def show_gray_img(img, title=""): plt.imshow(img) plt.title(title) plt.show() def corr(frame1, frame2): return np.sum(frame1 * frame2) / np.sqrt(np.sum(frame1 ** 2) * np.sum(frame2 ** 2)) def argmax_2d(array): h, w = array.shape max_idx = np.argmax(array) return max_idx // w, max_idx % w def argmin_2d(array): h, w = array.shape min_dix = np.argmin(array.flatten()) return min_dix // w, min_dix % w def ncc(template, scene): # 中心化 template = template - np.mean(template) scene = scene - np.mean(scene) template_h, template_w = template.shape scene_h, scene_w = scene.shape print(template_h, template_w, scene_h, scene_w) # 分窗 strides = scene.itemsize * np.array([scene_w, 1, scene_w, 1]) # print(strides) new_h = scene_h - (template_h - 1) new_w = scene_w - (template_w - 1) win_wrap_img = np.lib.stride_tricks.as_strided(scene, shape=(new_h, new_w, template_h, template_w), strides=strides) print(win_wrap_img.shape) corr_result = np.zeros((new_h, new_w)) for h in range(new_h): for w in range(new_w): corr_result[h][w] += corr(template, win_wrap_img[h][w]) # show_gray_img(corr_result) plot_3d(np.arange(new_w), np.arange(new_h), corr_result, "ncc corr result") max_local = argmax_2d(corr_result) return [*max_local, *template.shape] def hausdorff_dis(frame1, frame2): frame1_2 = np.min( ( abs(frame1.reshape([-1, 1]).repeat(frame2.shape[0], 1) - frame2) ).reshape([len(frame1.flatten()), -1]), axis=-1 ) frame2_1 = np.min( ( abs(frame2.reshape([-1, 1]).repeat(frame1.shape[0], 1) - frame1) ).reshape([len(frame2.flatten()), -1]), axis=-1 ) return np.max((np.max(frame1_2), np.max(frame2_1))) def hausdorff(template, scene): template = cv2.Canny(template, 150, 250) scene = cv2.Canny(scene, 150, 250) show_gray_img(scene) show_gray_img(template) stride = 3 template_h, template_w = template.shape scene_h, scene_w = scene.shape strides = scene.itemsize * np.array([scene_w * stride, 1 * stride, scene_w, 1]) new_h = (scene_h - template_h) // stride + 1 new_w = (scene_w - template_w) // stride + 1 win_wrap_img = np.lib.stride_tricks.as_strided(scene, shape=(new_h, new_w, template_h, template_w), strides=strides) temp_x, temp_y = np.where(template > 0) dis_result = np.zeros((new_h, new_w)) print(win_wrap_img.shape) for h in range(new_h): for w in range(new_w): win_x, win_y = np.where(win_wrap_img[h][w] > 0) dis_result[h][w] += np.max([hausdorff_dis(temp_x, win_x), hausdorff_dis(temp_y, win_y)]) plot_3d(np.arange(new_w), np.arange(new_h), dis_result, "hausdorff dis result") min_local = argmin_2d(dis_result) print(min_local) return [(min_local[0] - 1) * stride, (min_local[1] - 1) * stride, *template.shape] def hausdorff_distance_trans(bk_edge): row, col = bk_edge.shape min_dis = (row + col) * np.ones((row, col)) a, b = np.where(bk_edge == 255) for i in range(row): for j in range(col): min_dis[i, j] = np.min(abs(a - i) + abs(b - j)) return min_dis def distance_transform(template, scene): template = cv2.Canny(template, 150, 250) scene = cv2.Canny(scene, 150, 250) show_gray_img(scene) print(np.max(scene)) scene = hausdorff_distance_trans(scene) show_gray_img(scene) stride = 1 template_h, template_w = template.shape scene_h, scene_w = scene.shape strides = scene.itemsize * np.array([scene_w * stride, 1 * stride, scene_w, 1]) new_h = (scene_h - template_h) // stride + 1 new_w = (scene_w - template_w) // stride + 1 win_wrap_img = np.lib.stride_tricks.as_strided(scene, shape=(new_h, new_w, template_h, template_w), strides=strides) dis_result = np.zeros((new_h, new_w)) temp_x, temp_y = np.where(template > 0) print(win_wrap_img.shape) for h in range(new_h): for w in range(new_w): # dis_result[h][w] += corr(template, win_wrap_img[h][w]) # win_x, win_y = np.where(win_wrap_img[h][w] > 0) # dis_result[h][w] += np.max([hausdorff_dis(temp_x, win_x), hausdorff_dis(temp_y, win_y)]) dis_result[h][w] += np.mean(template * win_wrap_img[h][w]) plot_3d(np.arange(new_w), np.arange(new_h), dis_result, "distance_transform corr result") min_local = argmin_2d(dis_result) # print(min_local) return [(min_local[0] - 1) * stride, (min_local[1] - 1) * stride, *template.shape] def draw_rectangle(locals, scene, title=""): plt.imshow(scene, cmap='gray') plt.title(title) ax = plt.gca() # 默认框的颜色是黑色，第一个参数是左上角的点坐标 # 第二个参数是宽，第三个参数是长 names = ["Template_1.jpg", "Template_2.jpg"] for indx, local in enumerate(locals): ax.add_patch(plt.Rectangle((local[1], local[0]), local[2], local[3], color="blue", fill=False, linewidth=1)) ax.text(local[1], local[0], names[indx], bbox={'facecolor': 'blue', 'alpha': 0.5}) plt.title(title) plt.show() def find_position(templates, scene, function, title): scene_r = scene.shape[0] locals = [] for template in templates: locals.append(function(template, scene)) print(locals) print("左下角为(0,0)情况下,模版1和2目标框的左上角坐标分别为:\n", [[scene_r - item[0], item[1]] for item in locals]) draw_rectangle(locals, scene, title=title) if __name__ == '__main__': img_home = r"./data/template_matching/" template_test_1 = load_img(os.path.join(img_home, "Template_1.jpg")) template_test_2 = load_img(os.path.join(img_home, "Template_2.jpg")) scene_test = load_img(os.path.join(img_home, "Scene.jpg")) start = time.time() find_position([template_test_1, template_test_2], scene_test, ncc, "ncc") print("ncc耗时:", time.time() - start) start = time.time() find_position([template_test_1, template_test_2], scene_test, hausdorff, "hausdorff") print("hausdorff耗时:", time.time() - start) start = time.time() find_position([template_test_1, template_test_2], scene_test, distance_transform, "distance_transform") print("distance_transform+hausdorff耗时:", time.time() - start)