Image Classification Based on Codebook Generation using New Clustering Approach and Distinctive Feature Selection

Nowadays, in research under image processing domain, image retrieval and image classification have been a demanding research field. During the development of an effective classification system, selection of low-level visual features is an important step while selecting those features, two main problems are arises. First, in the feature extraction process, a huge number of key-points are situated in the image. Second, from the extracted features, most of the key-points are arise in background regions, which do not give the effective results in the classification process. In order to decrease the inverse effects of these limitations, In order to reduce those limitations, there is a classification system based on class specific codebook generation using Self Organizing Map (SOM) and distinctive feature selection method. But it is very complex process for large numbers of images from many categories. So we propose a new classification system based codebook generation using new clustering method along with a novel feature selection in the bag-of-words model. Then evaluate our classification system by compare the results with classification methods of K Nearest Neighbor, Support Vector Machine and Naive Bayes. We used Caltech101 image Dataset for experiments which shows that our classification system give the significant improvement in classification.