Question

Hi I have a project which related to the content based image retrieval. I have a doubt about what criteria should be taken for this project.when I searched this there I got two methods Feature extraction and image segmentation and region manipulation. Which method is more efficient?

Answer 1

Neither. Answer to this question will require some researrch on yor side. Image recognition is usually done via following steps

• some preprocessing to filter out unnecessary features ( depends on context, for text recognition you would binarise images , for face recognition may be made them grayscale… )
• finding region of interest ( in case of text recognition you would try to locate text lines / glyphs , in case of faces apply some haar cascade to find where faces are – also depends on your image set )
• feature extraction ( there is a ton of different methods, for some shapes invariant moments like Hu are good solution, for faces you could get better results with Zernike moments – they are based on polar coordinates )
• matching ( easier part: you have got some feature vector, typically doubles, and now you have to match it to other feature vector / class of vectors – here you go with cluster analysis, there are also different techniques )

You may see it all together in action in JavaOCR project ( text recognition for android ) – but please compile it from source as release is pretty old ):

http://sourceforge.net/projects/javaocr/