sift - Capturing macro and micro geometry using image features -


i extracting image features texture identification. want extract image features 2d image. these features should capture micro macro geometry of image. have tried find out algorithm can give me above result. tried sift captures details of macro geometry, shape , edge details. on other hand have tried various versions of lbp (like cs-lbp, lnbp etc) can capture pixel differences , not overall scheme of things. technique tried using glcm different distance , orientation values . still there problem resulting histogram large , time taken comparison large. additionally, results not also. classification rate svm 48%. can guide me in finding generic algorithm can capture pixel level differences overall texture pattern.

why need generic algorithm? have:

  1. local non parametric features (like lbp) capture micro
  2. macro features (like corner detectors, sift)
  3. overall geometry invariant statistics (like histograms, fourier transform etc)

just calculate features, concatenate them in single vector , vector result wanted. don't need single smart algorithm. key success combination (concatenation) of many "simple" algorithms.

if have speed problems (your vector of features long) try using pca before svm. reduce size of vector in such way loose "minimal amount of information". pca can reduce length of vector factor of 10 easily. pca running time relatively fast (just multiplying 2d matrix).

if still have low recognition rates when sure used features try itml. itml used after pca , multiplication 2d matrix. itml way "normalize affection of each feature , cancel out affect of feature's large or small range of values".

typically if use features, run pca, itml , svm - should results on difficult computer vision tasks


Comments

Popular posts from this blog

angularjs - ADAL JS Angular- WebAPI add a new role claim to the token -

php - CakePHP HttpSockets send array of paramms -

node.js - Using Node without global install -