Content-Based Image Retrieval Using Features Extracted From Halftoning-Based Block Truncation Coding

This paper presents a technique for content-based image retrieval (CBIR) by exploiting the advantage of low-complexity ordered-dither block truncation coding (ODBTC) for the generation of image content descriptor. In the encoding step, ODBTC compresses an image block into corresponding quantizers and bitmap image. Two image features are proposed to index an image, namely, color co-occurrence feature (CCF) and bit pattern features (BPF), which are generated directly from the ODBTC encoded data streams without performing the decoding process. The CCF and BPF of an image are simply derived from the two ODBTC quantizers and bitmap, respectively, by involving the visual codebook. Experimental results show that the proposed method is superior to the block truncation coding image retrieval systems and the other earlier methods, and thus prove that the ODBTC scheme is not only suited for image compression, because of its simplicity, but also offers a simple and effective descriptor to index images in CBIR system.

I. Introduction

Asignificant amount of research efforts have been devoted in addressing the Content Based Image Retrieval (CBIR) problem [11]–[22], [26]–[71]. An image retrieval system returns a set of images from a collection of images in the database to meet users’ demand with similarity evaluations such as image content similarity, edge pattern similarity, color similarity, etc. An image retrieval system offers an efficient way to access, browse, and retrieve a set of similar images in the real-time applications. Several approaches have been developed to capture the information of image contents by directly computing the image features from an image as reported in [15]–[20]. In [21], the image feature is simply constructed in DCT domain. An improvement of image retrieval in DCT domain is presented in [22], in which the JPEG standard compression (excluding the entropy coding) is involved to generate the image feature. Some attempts have been addressed to describe the visual image content [26]–[28], [30]–[34]. Most of them are dealing with the MPEG-7 Visual Content Descriptor, including the Color Descriptors (CD), Texture Descriptor (TD), and Shape Descriptor (SD) to establish the international standard for the CBIR task. This standard provides a great advantage in the CBIR research field, in which some important aspects such as sharing the image feature for benchmark database, comparative study between several CBIR tasks, etc., become relatively easy to be conducted using these standard features. The standard also offers a great benefit in the distributed system, in which the image content descriptor can be remotely modified by the user. In this scenario, the original image is not necessary transferred over different locations, but only the image descriptor is required for modification and recalculation. A new type of CBIR approach is presented in [65], in which the spatial pyramid and orderless bag-of-features image representation were employed for recognizing the scene categories of images from a huge database. This method offers a promising result and outperforms the former existing methods in terms of the natural scene classification. The method in [66] presented the holistic representation of spatial envelop with a very low dimensionality for representing the scene image. This approach presented an outstanding result in the scene categorization. The method in [67] proposed a new approach for image classification with the receptive field design and the concept of over-completeness methodology to achieve a preferable result. As reported in [67], this method achieved the best classification performance with much lower feature dimensionality compared to that of the former schemes in image classification task.