Hybrid Approach for Image Retrieval Using Local Directional Texton Histograms (LDTH) and Color Features
Main Article Content
Abstract
The texton based methods played a significant role in retrieving the relevant objects from the given data bases. The textons in the literature are mostly derived on a 2*2 grid. This paper attempted to derive the 1*3 texton patterns on 3*3 windows. These textons are derived on different directions. This paper initially divides the image into an overlapped grids of same 3*3. The four directional triangular textons and four line directional textons are computed on each window. A ternary pattern is computed on each of the texton grid and corresponding texton unit is derived. The texton unit replaces the center pixel. This process transforms the image into 2 images, i.e., one with local triangular textons units and the other with local line texton units, each of which ranging from 0 to 80. A combined histogram named local directional texton histogram is derived from these transformed images and integrated with color histograms. This process derives the image features. These features are used for CBIR and the results indicate high performance when compared with other texton based methods.