OpenCV中如何在没有Sobel滤波器的情况下过滤线路
我正在尝试仅使用线性滤波器来检测线条。我的第一次尝试是像这样旋转内核，但行不通：
kernel = zeros((13,13))
kernel60 = zeros((13,13))
kernel[4] = [0,0,0,0,-1,-1,-1,-1,-1,0,0,0,0]
#kernel[5] = [0,0,0,0,0]
kernel[6] = [0,0,0,0,2,2,2,2,2,0,0,0,0]
#kernel[7] = [0,0,0,0,0]
kernel[8] = [0,0,0,0,-1,-1,-1,-1,-1,0,0,0,0]
rotate60 = zeros((2,3))
GetRotationMatrix2D((6,6),60,1, rotate60)
WarpAffine(kernel,kernel60,rotate60,CV_WARP_FILL_OUTLIERS, ScalarAll(0))
之后我准备了一个内核，它是两个 Sobel 内核（可操纵滤波器）的线性组合。这可行，但我最好喜欢非索贝尔内核，类似于第一次尝试。索贝尔内核有什么替代品吗？
Sobel Kernel组合：
kernel_x[0] = [-1,0,+1]
kernel_x[1] = [-1,0,+1]
kernel_x[2] = [-1,0,+1]
kernel_y[0] = [-1,-1,-1]
kernel_y[1] = [0,0,0]
kernel_y[2] = [+1,+1,+1]
normal_theta = radians(-30)
kernel = multiply(cos(theta),kernel_x) + multiply(sin(theta),kernel_y)
然后过滤：
Filter2D(src,dst,kernel)
我在Windows机器中使用Python和numpy。
I'm trying to detect lines just with linear filters. My first try was rotate a kernel like this but wouldn't work:
kernel = zeros((13,13))
kernel60 = zeros((13,13))
kernel[4] = [0,0,0,0,-1,-1,-1,-1,-1,0,0,0,0]
#kernel[5] = [0,0,0,0,0]
kernel[6] = [0,0,0,0,2,2,2,2,2,0,0,0,0]
#kernel[7] = [0,0,0,0,0]
kernel[8] = [0,0,0,0,-1,-1,-1,-1,-1,0,0,0,0]
rotate60 = zeros((2,3))
GetRotationMatrix2D((6,6),60,1, rotate60)
WarpAffine(kernel,kernel60,rotate60,CV_WARP_FILL_OUTLIERS, ScalarAll(0))
After that I prepared a kernel that's a linear combination from two Sobel kernels (steerable filters). This works but I would better like a non-sobel kernel, similar to the first try. Any alternative to the sobel kernels?
Sobel Kernel combination:
kernel_x[0] = [-1,0,+1]
kernel_x[1] = [-1,0,+1]
kernel_x[2] = [-1,0,+1]
kernel_y[0] = [-1,-1,-1]
kernel_y[1] = [0,0,0]
kernel_y[2] = [+1,+1,+1]
normal_theta = radians(-30)
kernel = multiply(cos(theta),kernel_x) + multiply(sin(theta),kernel_y)
Then filtering:
Filter2D(src,dst,kernel)
I use Python and numpy in a Windows machine.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
您可以使用 Canny 算法进行边缘检测（无论如何都使用 Sobel），使用霍夫变换进行直线检测。在 Canny 之前执行模糊可以帮助消除异常线。这是经典的方法。您可以使用 OpenCV 来实现这两个部分。
请参阅以下内容：
http://en.wikipedia.org/wiki/Hough_transform
http://en.wikipedia.org/wiki/Canny_edge_detector
以下是 OpenCV 实现的文档：
请参阅 cvHoughLines* 函数，其中有示例代码
You can use Canny algorithm for edge detection (which uses Sobel anyway) and Hough transform for line detection. Performing blur before Canny can help eliminate outlier lines. This is the classic approach. You can use OpenCV that implements both parts.
See the following:
http://en.wikipedia.org/wiki/Hough_transform
http://en.wikipedia.org/wiki/Canny_edge_detector
Here is the documentation for OpenCV implementation:
http://opencv.willowgarage.com/documentation/cpp/imgproc_feature_detection.html
see the cvHoughLines* functions there are sample code