Feature encoding is a crucial step in BOW image representation. The standard BOW model assigns each image feature to the nearest visual-word without making a distinction between the features that are assigned to the same words. This hard feature assignment leads to high quantization errors and degrades the learning capacity of the classifiers in image classification. We propose a fuzzy feature encoding approach to overcome the uncertainty problem in BOW through assigning each image feature to the visual-words with some membership degrees. We employ two classification techniques, Naive Bayesian and SVM, to evaluate the effect of the fuzzy assignment in image classification. Experiments conducted on image datasets show that fuzzy feature encoding significantly improves the classification accuracy.

The bag-of-words (BOW) has become a popular image representation model with successful implementations in visual analysis. Although the original model has been improved in several ways, the utilization of the Fuzzy Set Theory in BOW has not been investigated thoroughly. This paper presents a fuzzy feature encoding approach to address the problems associated with the hard and soft assignments of image features to the visual-words. Our encoding method assigns each image feature to only the first and second closest words in the codebook to overcome the word-uncertainty problem. Moreover, we introduce a new word-weighting scheme for image categories based on image histograms. Experiments conducted on some image datasets show that both methods increase the BOW performance in content based image retrieval.

Local image features, which are robust to scale, view, and orientation changes in images, play a key factor in developing effective visual classification systems. However, there are two main limitations to exploit these features in image classification problems: 1) a large number of key-points are located during the feature detection process, and 2) most of the key-points arise in background regions, which do not contribute to the classification process. In order to decrease the inverse effects of these limitations , we propose a new codebook generation approach through employing a new clustering method that generates class-specific codebooks along with a novel feature selection method in the bag-of-words model. We evaluate the performance of different classification techniques including Naive Bayesian, k-NN, and SVM on distinctive features. Experiments conducted on PASCAL Visual Object Classification collections have shown that the class-specific codebooks along with distinctive image features can significantly improve the classification performances.

Support Vector Machines (SVMs) have been extensively used for visual object classification to bridge the semantic gap between the low level features and high level concepts. SVM treats each training input equally during the construction of its decision surface which results in poor learning machines if training data include outliers. In this paper, a novel fuzzy visual object classification approach utilizing Self-Organizing Maps (SOMs) in SVM is proposed. The experimental results show the effectiveness of the proposed Fuzzy SVM compared to the traditional SVM.