python-tensorflow-opencv《計算機視覺的深度學習實踐》代碼筆記

1、ython-tensorflow-opencv計算機視覺的深度學習實踐代碼筆記python-tensorflow-opencv計算機視覺的深度學習實踐代碼筆記01 課程概述tf_test:# 創(chuàng)建2個矩陣，前者12列，后者21列，然后矩陣相乘：matrix1 =tf.constant(3,3)# 上邊的操作是定義圖，然后會話Session去計算：with tf.Session() as sess:print(- )print(matrix1)Tensor(Const_2:0, shape(1, 2), dtypeint32)opencv_test:import cv2 as cv#釋放窗cv.

2、destroyAllWindows()02圖像預處理f1.py程序：import cv2img =cv2.imread(D:/pictures/opencv.png,0) #直接讀為灰度圖像for i in range(2000): #添加點噪聲blur_1 =cv2.GaussianBlur(img,(5,5),0)blur_2 =cv2.medianBlur(img,5)plt.subplot(1,3,1),plt.imshow(img,gray),plt.title(img)#默認彩，另種彩bgrplt.subplot(1,3,2),plt.imshow(blur_1,gray),plt

3、.title(blur1)plt.subplot(1,3,3),plt.imshow(blur_2,gray),plt.title(blur2)plt.show()f1.py結(jié)果：f2.py程序：import cv2 as cvfrom matplotlib import pyplot as pltimg =cv.imread(D:/pictures/dave.png,0)plt.subplot(2,2,4),plt.imshow(sobely,cmap =gray)plt.title(Sobel Y), plt.xticks(), plt.yticks()plt.show()f2.py結(jié)果：

4、fft.py程序：import cv2 as cvplt.subplot(121),plt.imshow(img, cmap =gray)plt.title(Input Image), plt.xticks(), plt.yticks()plt.subplot(122),plt.imshow(magnitude_spectrum, cmap =gray)plt.title(Magnitude Spectrum), plt.xticks(), plt.yticks()plt.show()fft.py結(jié)果：clahe.py程序：import cv2import matplotlib.pyplot

5、as pltimg =cv2.imread(D:/pictures/timg.jpg,0) #直接讀為灰度圖像res =cv2.equalizeHist(img)#直接直圖均衡化clahe =cv2.createCLAHE(clipLimit2,tileGridSize(10,10)cl1 =clahe.apply(img)# 限制對度適應直圖均衡plt.show()clahe.py結(jié)果：filter_1.ipynb程序：kernel =np.ones(5,5),np.float32)/25dst =cv.filter2D(img,-1,kernel)plt.show()kernelfilte

6、r_1.ipynb結(jié)果：array(0.04, 0.04, 0.04, 0.04, 0.04,0.04, 0.04, 0.04, 0.04, 0.04,0.04, 0.04, 0.04, 0.04, 0.04,0.04, 0.04, 0.04, 0.04, 0.04,0.04, 0.04, 0.04, 0.04, 0.04, dtype=float32)03圖像特征與描述imagestiching.py程序：def stitch(self, images, ratio0.75, reprojThresh4.0,showMatchesFalse):#獲取輸圖# 匹配兩張圖的所有特征點，返回匹配結(jié)

7、果# 如果返回結(jié)果為空，沒有匹配成功的特征點，退出算法if M is None:return None# 否則，提取匹配結(jié)果# 將圖進視變換，result是變換后圖result =cv2.warpPerspective(imageA, H, (imageA.shape1 +imageB.shape1, imageA.shape0)# 將圖傳result圖最左端# 檢測是否需要顯圖匹配# 成匹配圖vis =self.drawMatches(imageA, imageB, kpsA, kpsB, matches, status)# 返回結(jié)果return (result, vis)# 返回匹配結(jié)果r

8、eturn resultdef detectAndDescribe(self, image):# 將彩圖轉(zhuǎn)換成灰度圖gray =cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)descriptor =cv2.xfeatures2d.SIFT_create()# 檢測SIFT特征點，并計算描述(kps, features) =descriptor.detectAndCompute(image, None)# 將結(jié)果轉(zhuǎn)換成NumPy數(shù)組kps =np.float32(kp.pt for kp in kps)# 返回特征點集，及對應的描述特征return (kps, f

9、eatures)def matchKeypoints(self, kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh):# 建暴匹配器matcher =cv2.DescriptorMatcher_create (BruteForce)# 使檢測來圖的SIFT特征匹配對，K=2rawMatches =matcher.knnMatch(featuresA, featuresB, 2)# 當最近距離跟次近距離的值于ratio值時，保留此匹配對if len(m) =2 and m0.distance m1.distance *ratio:# 存儲

10、兩個點在featuresA, featuresB中的索引值matches.append(m0.trainIdx, m0.queryIdx)# 當篩選后的匹配對于4時，計算視變換矩陣# 獲取匹配對的點坐標# 計算視變換矩陣# 返回結(jié)果# 如果匹配對于4時，返回None# 聯(lián)合遍歷，畫出匹配對for (trainIdx, queryIdx), s) in zip(matches, status):# 當點對匹配成功時，畫到可視化圖上if s =1:# 畫出匹配對return vis# 讀取拼接圖# 把圖拼接成全景圖stitcher =Stitcher()(result, vis) =stitche

11、r.stitch(imageA, imageB, showMatchesTrue)# 顯所有圖cv2.destroyAllWindows()imagestiching.py結(jié)果：harris_corner.py程序：import numpy as npimport cv2 as cvfilename =./chessboard.pngimg =cv.imread(filename)cv.imwrite(D:/dst.jpg,img)if cv.waitKey(0) &0 xff =27:cv.destroyAllWindows()harris_corner.py結(jié)果：surf.py程序：imp

12、ort numpy as npimport cv2 as cvimg =cv.imread(./butterfly.jpg,0)surf =cv.xfeatures2d.SURF_create(400)surf.setHessianThreshold(50000)kp, des =surf.detectAndCompute(img,None)cv.waitKey(0)cv.destroyAllWindows()surf.py結(jié)果：sift.py程序：import cv2 as cvimgcv.drawKeypoints(gray,kp,img)cv.destroyAllWindows()sif

13、t.py結(jié)果：orb.py程序：import cv2 as cvimg3 =cv.drawMatches(img1,kp1,img2,kp2,matches:20,None, flags2)plt.imshow(img3),plt.show()orb.py結(jié)果：laplacian_sharpen.py程序：# 為了減少計算的維度，因此將圖像轉(zhuǎn)為灰度圖img_gray =img.convert(L)# 得到轉(zhuǎn)換后灰度圖的像素矩陣img_arr =np.array(img_gray)h =img_arr.shape0 # w =img_arr.shape1 # 列# 拉普拉斯算銳化圖像，階微分new_img_arr =np.zeros(h, w) # 拉普拉斯銳化后的圖像像素矩陣for i in range(2, h-1):laplace_img_arr =np.zeros(h, w)