The Statistical Quantized Histogram Texture Features Analysis for Image Retrieval Based on Median and Laplacian Filters in the DCT Domain

An effective Content-Based Image Retrieval (CBIR) system is based on efficient feature extraction and accurate retrieval of similar images. Enhanced images by using proper filter methods can also, play an important role in image retrieval in a compressed frequency domain since currently most of the images are represented in the compressed format by using the Discrete Cosine Transformation ( DCT) blocks transformation. In compression, some crucial information is lost and perceptual information is left, which has significant energy requirement for retrieval in a compressed domain. In this paper, the statistical texture features are extracted from the enhanced images in the DCT domain using only the DC and first three AC coefficients of the DCT blocks of image having more significant information. We study the effect of filters in image retrieval using texture features. We perform an experimental comparison of the results in terms of accuracy on the basis of median, median with edge extraction and Laplacian filters using quantized histogram texture features in a DCT domain. Experiments on the Corel database using the proposed approach, give the improved results on the basis of filters; more specifically, the Laplacian filter with sharpened images gives good performance in retrieval of JPEG format images as compared to the median filter in the DCT frequency domain.