problem in Concatenating feature vectors from two different image.

Dear Sir/Madam,

I am working on feature level fusion, in which feature vectors extracted from two different images (one image is binary image and another is a gray scale image) are concatenated. Here the feature vectors are extracted using Principal component Analysis Technique. I normalized the feature vectors before concatenating them. When i measured the Euclidean distance between two different concatenated feature vectors gives me poor result. how to solve it?

[mitali]

could u please pass the code for normalizing the Eigen vectors

Answer 1

[Image Analyst]

I don't know how you normalized them. You'd need to normalize each feature by itself before adding to other features. For example if one feature was the number of pixels in a blob, and the other was the number of holes in a blob, then you can't just do sqrt(area1-area2)^2+(holes1-holes2)^2) because the number of pixels feature would totally swamp the number of holes feature making the number of holes feature essentially ignored completely.

PCA gives a new coordinate system so I don't really know what feature vectors you extracted from this new coordinate system - it could be anything, but you didn't say what. It could be the mean PC values, or the distance of one PC value to another (like percent of the way some colored material is from orange to aqua), or virtually anything.

[faa nad]

I have taken the Eigen vectors returned by PCA as the feature vectors from both the images(say face and iris images).As you said ,I normalized the features before concatenating them using z-score normalization method.when I measured Euclidean distance of the feature vector of two face images and two iris images separately ,and then adding their scores gives good result.but concatenation of iris and face feature vectors and measuring their distance gives poorer result

[Image Analyst]

There's not enough information for me to debug why, and if I did have enough information, I wouldn't have enough time to fix your application. So about all I can say is to keep working on it and I think eventually you'll get it. Good luck.