I am working on Low Resolution Face recognition to improve it's Recognition rate which involves Feature extraction. I intend to use GLCM and LBP for feature extraction + either SVM or HMM classifier. Can GLCM and LBP be a good feature extraction techniques ?? Please advise should there be a better techniques to yield a good result.
* Some of the texture feature-based methods work well even in case of low resolution images. LBP is one such texture feature which gives more local-information which can be very handy especially in the case of face recognition. Thus, it is being used by many researchers for facial image analysis.
*This paper is a good read : Article Local Binary Patterns and Its Application to Facial Image An...
* There are different versions of LBP that can be experimented for your application to understand which version would give better accuracy.
*Please note that the conventional LBP has a large feature vector length and thus computationally expensive.
* Also, others versions of LBP may be comparatively less sensitive to noise (which is more in low resolution images) . For eg: LTP is less sensitive to noise. The following paper is a good read:
Article Lecture Notes in Computer Science
*GLCM gives second order texture information. Using GLCM, statistical features of the image can be extracted (energy, correlation, homogeneity etc.). As compared to LBP, GLCM is less efficient in case of facial image analysis.
*There are two possible ways to combine GLCM and LBP.
a) By extracting LBP first, and then extracting GCLM features from the LBP image.
b) By concatenating the feature vectors obtained after extracting LBP and GLCM seperately.
* Check this paper for a related work:
Conference Paper A Novel Crowd Density Estimation Technique using Local Binar...
* Also, you can experiment with another efficient global texture feature: Gabor wavelet/ filter!!
Good luck!!
Would recognize class as face or No face in the image?
In this case, Viola et Jones algo is a good choice.
If you try to recognize a particular people inside the image this kind of descriptor give great results : Article Generalized Fourier Descriptors with Applications to Objects...
Good luck!
Combing many feature may not be efficient in terms of complexity. You should try to use deep CNN feature which can extract all hidden features from an image including edges, color, texture etc.
I have fused color and RLBP in my paper but these days CNN is more effective than handcrafted features.
Article Integrating salient colors with rotational invariant texture...
U can use my method for texture faeture extraction. It provide comparetivley better results to LBP..
@Mohd Dilshad Ansari. Thanks so much for your suggestions. Permit me to ask, what method ?? I will appreciate.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India