some questions

[weihangChen]

Please help , Cesar

1. when I compare the source code between project MachineLearning.Handwriting(SVM) and Imaging.Classification(BOW), the way to extract feature vector is very different, the first one just extract the pixel data as double[] while the second one use binarysplit/kmeans to compute the featurevector, I wonder in which case the kmeans should be used.
2. I am extracting contours from a bigger image, so the contour letters are in various size, ex. Height33, Width24....(lets assume the training images are in size of 64*64), the computed result have a very low matching rate, I guess this problem is due to the image size, what should I do ?