[Study Notes] Week1_Convolutional Neural Networks_More Edge Detection

1. Vertical edge detection: the difference between light-to-dark edges and dark-to-light edges

    If you don't care about these, you can take the absolute value of the convolution result

    

2. Horizontal edge detection

    Different convolution kernels can detect edges in different directions

    

3. Other filters:

    1) Sobel Filter

    

    2) Scharr Filter ( Scharr filter )

        

    Note: The above are all vertical edge detection filters, and the horizontal edge detection filter can be obtained by rotating the filter by 90°

4. The elements of the filter can also be used as parameters to learn through backpropagation (for convolution operations), and even a filter for detecting arbitrary edge angles can be learned, and the effect is often better than manually set filters.

    

Guess you like

Origin http://43.154.161.224:23101/article/api/json?id=325524045&siteId=291194637