Opencv Extract最大的感兴趣区域

时间:2017-05-01 11:14:15

标签: python image opencv opencv3.0 opencv-contour

我有几张图片如下所示: 图像主要是在白色背景上。

在白色(大多数)背景上布置了多件衣服。 enter image description here enter image description here

我尝试使用opencv连接组件检测两件衣服。 试图采取最大的两个连接组件,不幸的是,我失败了。

我相信这是可能的,但由于我是opencv的新手,有人可以对以下图像中可以检测多件衣服的方法有所了解吗?

感谢任何帮助

我在python中试过的代码:

#Read the image and conver to grayscale
img = cv2.imread('t.jpg' , 0)
#Applt the median filter on the image 
#med = cv2.medianBlur(image,5)    # 5 is a fairly small kernel size
#Apply an edge detection filter 

laplacian = cv2.Laplacian(img,cv2.CV_64F)

laplacian = laplacian.astype(np.uint8)
ret,thresh1 = cv2.threshold(laplacian,127,255,cv2.THRESH_BINARY)
src = thresh1
src  = np.array(src, np.uint8)
ret, thresh = cv2.threshold(src,10,255,cv2.THRESH_BINARY)
# You need to choose 4 or 8 for connectivity type
connectivity =8
# Perform the operation
output = cv2.connectedComponentsWithStats(thresh, connectivity, cv2.CV_32S)
# Get the results
# The first cell is the number of labels
num_labels = output[0]
# The second cell is the label matrix
labels = output[1]
# The third cell is the stat matrix
stats = output[2]
# The fourth cell is the centroid matrix
centroids = output[3]
src = cv2.cvtColor(src,cv2.COLOR_GRAY2RGB)
for stat in stats:
    x , y ,w , h ,a = stat
    cv2.rectangle(src,(x,y),(x+w,y+h),(0,0,255),2)
    # write original image with added contours to disk
    #cv2.imwrite('contoured.jpg', image)
cv2.imshow("Image", src)
#cv2.waitKey(0)
cv2.waitKey(0)
cv2.destroyAllWindows()

上述代码的输出

enter image description here

NB :即使我可以提取给定图像中的最大对象,也很好。

1 个答案:

答案 0 :(得分:1)

这是一个非常简单的方法,只需使用图像阈值处理并找到轮廓来提取第二张图片中最大的服装项目。要获得其他项目,您只需调整阈值以使其不被消除,然后您将搜索轮廓。不是最好的解决方案,但这是一个开始。

img = cv2.imread('t.jpg' , 0) # import image as grayscale array

# threshold image
img_b = cv2.GaussianBlur(img, (13, 13), 2)
ret, img_th = cv2.threshold(img_b, 40, 255, cv2.THRESH_BINARY_INV)
# find contours
(_,cnts,_) = cv2.findContours(img_th.copy(), cv2.RETR_TREE, 
cv2.CHAIN_APPROX_SIMPLE)
print(str(len(cnts))+' contours detected')

# find maximum area contour
area = np.array([cv2.contourArea(cnts[i]) for i in range(len(cnts))]) # 
list of all areas
maxa_ind = np.argmax(area) # index of maximum area contour

plt.figure(figsize=(10,4))
plt.subplot(1,3,1)
plt.imshow(img_b)
plt.title('GaussianBlurr')
plt.subplot(1,3,2)
plt.imshow(img_th)
plt.title('threshold')
plt.subplot(1,3,3)
xx = [cnts[maxa_ind][i][0][0] for i in range(len(cnts[maxa_ind]))]
yy = [cnts[maxa_ind][i][0][1] for i in range(len(cnts[maxa_ind]))]
ROI.append([min(xx),max(xx),min(yy),max(yy)])
plt.imshow(img)
plt.plot(xx,yy,'r',linewidth=3)
plt.title('largest contour')

此代码生成以下图像:

enter image description here