SVM Data

[saro...@gmail.com]

Hello,

What method do you would advice to create data for SVM ?
Do you cut image by sector and affect a value to the average sum of each pixel of the zone, or we should use more accurate algorithm ?

Thank you.

[César]

Hello there,

Well, this way could be feasible; however there are also other ways to obtain feature vectors from images. One possible way, for example, is to extract feature vectors using Histogram of Oriented Gradients (which are also available on the framework: http://accord-framework.net/docs/html/AllMembers_T_Accord_Imaging_HistogramsOfOrientedGradients.htm ). Other ways include extracting features using Local Binary Patterns (i.e. http://accord-framework.net/docs/html/T_Accord_Imaging_LocalBinaryPattern.htm ) and Haralick's features (mostly for texture analysis, using http://accord-framework.net/docs/html/AllMembers_T_Accord_Imaging_Haralick.htm ).

Another more popular, but potentially slower way is to create Bag of Visual Words using a feature extraction algorithm and a clustering algorithm, such as is shown in the Image Classification sample application. Hope it helps!

Best regards.

Cesar

[saro...@gmail.com]

Thank, these solutions look very interesting.