scikit learn + random forest segmentation

1
votes

I am using random forest function in scikit learn for segmentation of image. However i am not able to create an input to the function clf.fit(X,Y). X is a training matrix of (n_samples,n_features), Y is a target matrix of (n_samples,) in which the last attribute is the labels of the target class. I want to train 50 images with four-five features, like HOG features,RGB features,f17 filter and texon map.

Can anyone please help me create the matrix X and Y. Thanks!

1
scikit-learnrandom-forestimage-segmentation
Are you using a RandomForestClassifier to classify pixel-wise for your segmentation? You need to be more specific about what your doing, and add some example code. – Matt Hancock
Just to give you an idea. You usually need 1000s of images to train something really simple like recognizing black digits on a black background. With only 50 images you won't get meaningful results. And a feature for an image is one pixel. Meaning a 50x50 pixel with rgb has 750 features. – Zephro

1 Answers

1
votes

If you're just classifying each pixel on its own, just put the rgb values into X and the segment as Y.

Something likely better though would be to pick a small region around the pixel, and either use near neighbors rgb values - all concated from left to right in each row, or the distributions of each of r,g,b within a local region.