Why do I must use Sobel Operator?

Question

As I've recently read some journals and pdfs about Neural Network. And i anchored my mind to an article about "Handwriting Recognition Using Neural Network". For addition, I'm studying Backpropagation. And my question is "Why do the author recommend the reader to use Sobel Operator as a Preprocessing Technique instead of plainly process the handwriting to the neural network ?

Answer 1

Handwriting is composed by strokes, usually the strokes are filled with a solid color: with these assumption a stroke is well described by its edges while its color or the background color is not so useful in describing it. The edges are basically described by boolean values: a pixel is white when it is part of an edge, black otherwise.

The grey level of the output image of the Sobel filter is proportional to the strength of the edges and so it can be used to describe the edges and then to describe the strokes, or, to "capture" the essence of the handwritten letters and digits.

If you feed the network with the raw image of an handwritten letter or digit there is a risk that the network learns some not so useful features like the full grey levels that do not describe very well the letter or the digit.

Some authors claim that the Sobel filter gives the edges in a brightness invariant way even if I think it is not strictly true.

Just out of curiosity, what is the paper where the author recommends the reader to use a Sobel filter?