Global and Local Image Descriptors for Content Based Image Retrieval and Object Recognition

This article develops two descriptors global and local for colored images. The first allows to do image retrieval while the second is applied to object recognition. First we combine color and texture attributes in a unique framework. Our approach is based on the physical properties of light bouncing from a scene. Furthermore, it is claimed that the distribution of Gabor filter outputs can also be provided by a Rayleigh. Filter.Thus, we deduce the RayleighGaussian framework from the Gabor-Gaussian one. Then we compute a descriptor based on statistical measures performed on Rayleigh-Gaussian features. Secondly, we propose a new detection function for interest points and a new characterization of such points. The detection function extends the Harris Laplace method. Our function extracts features invariant to rotation and scale from color images. We use the second moment as a basis of corner decision. Around each feature point we compute a texture descriptor using the Gabor filter. As an evaluation of this new class of interest points, we have implemented a recognition method of 3D objects by indexing a data base on object views; this method is similar to the Lowe method, except that that SIFT features and descriptors are replaced by the ones proposed in our approach.