图像处理：视网膜图像视盘分割效果优化技术求助

我是图像处理领域的新手，目前正在开展视网膜图像视盘分割项目。已完成ROI选取、红通道提取、滤波、阈值处理及形态学操作，相关内容如下：

输入图像 #

已实现的代码模块 #

ROI选取代码 #

from numpy.ma.core import ndim
def getROI(image):
    b,g,r = cv2.split(image)
    b = cv2.Blur(b,(15,15),0)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(15,15))
    b = ndimage.grey_opening(b,structure=kernel)    
    (minVal, maxVal, minLoc, maxLoc) = cv2.minMaxLoc(b)

    x0 = int(maxLoc[0])-150
    y0 = int(maxLoc[1])-150
    x1 = int(maxLoc[0])+150
    y1 = int(maxLoc[1])+150
    
    return image[y0:y1,x0:x1]

plt.imshow(image)
plt.show()

红通道提取代码 #

def rgb2Red(img):
    b,g,r = cv2.split(img)
    return r
def rgb2Gray(img):
    return cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

滤波处理代码 #

def clahe_image(image, clipLimit = 1.0, channels = 'a'):
    old_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    R,G,B = cv2.split(old_image)
    
    clahe = cv2.createCLAHE(clipLimit=clipLimit, tileGridSize=(8,8))
    
    if channels is 'g': 
        G = clahe.apply(G)
    if channels is 'b': 
        B = clahe.apply(B) 
    if channels is 'r': 
        R = clahe.apply(R) 
    if channels is 'a':
        G = clahe.apply(G)
        B = clahe.apply(B)
        R = clahe.apply(R)

    clahe_image = cv2.merge((R, G, B))

    return clahe_image

def preprocess(image):
    gray_blur = cv2.GaussianBlur(image, (3,3), 0);
    gray = cv2.addWeighted(image, 1.5, gray_blur, -0.5, 0, image);
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(31,31));
    gray = ndimage.grey_closing(gray,structure=kernel);
    gray = cv2.equalizeHist(gray);
    gray = clahe_image(gray);
    return gray

阈值处理代码 #

def optimal_threshold(h,t):
    h1 = h[:t]
    h2 = h[t:]
    m1 = (h1*np.arange(0,t)).sum()/h1.sum()
    m2 = (h2*np.arange(t,len(h))).sum()/h2.sum()
    t2 = int(np.round((m1+m2)/2))
    print(m1,m2,t2)
    if( t2 != t ) : return optimal_threshold(h,t2)
    return t2

h,bins = np.histogram(image,range(257))
t = optimal_threshold(h,255)
print(t)

ret, final_img = cv2.threshold(image,250,255,cv2.THRESH_BINARY)

形态学操作代码 #

img = final_img
kernel = np.ones((3,3), np.uint8)  
img_erosion = cv2.erode(img, kernel, iterations=1)  
img_dilation = cv2.dilate(img, kernel, iterations=4)  
plt.imshow(img_dilation, cmap='gray')   

当前分割结果与期望效果 #

当前分割结果 #

期望真值分割效果 #

求助问题 #

如何实现与真值图效果一致的视盘完美分割？添加血管去除步骤是否可行？

内容的提问来源于stack exchange，提问作者SHAFIRA