DETECTION OF PORNOGRAPHIC IMAGES USING BAG-OF-VISUAL-WORDS AND ARCX4 OF RANDOM MULTINOMIAL NAIVE BAYES

Abstract

The paper presents a novel approach to detect pornographic images. At the pre-processing step, we propose to use the Scale-invariant feature transform method (SIFT) which is locally based on the appearance of the object at particu­lar interest points, invariant to image scale, rotation and also robust to changes in illumination, noise, occlusion. And then, the representation of the image that we use for classification is the bag-of-visual-words (BoVW), which is constructed from the local descriptors and the counting of the occurrence of visual words in a histogram like fashion. The pre-processing step brings out datasets with a very large num­ber of dimensions. And then, we propose a new algorithm called Arcx4 of random multinomial naive Bayes (Arcx4-rMNB) that is suited for classifying very-high­-dimensional datasets. We do setup experiment with two real datasets to evaluate performances. Our approach has achieved an accuracy of 91.75% for a small dataset and 87.93% for other large one.