Research And Hardware Design Of MPEG-7 Compact Colour Descriptor

With the development of the computer and internet technology, image andaudio-visual information are more and more available. This makes the use ofdesired multimedia information more difficult, as more and more data has to besearched and lack of efficient retrieval methods. So the content-based search andretrieval technology emerges as the times requirement, which describes andsearches multimedia content by their own features. To adapt the requirement of the rapid increase of multimedia content and thedevelopment of content-based multimedia technology, it needs to specifydescriptions of multimedia object. In 1997, the ISO MPEG Group initiated the"MPEG-7 Multimedia Description Language" work item. The target of this activitywas to issue an international MPEG-7 Standard, defining standardized descriptionsand description systems that allow users or agents to search, identify, filter andbrowse audio-visual content. In September 2001, MPEG-7 was completed andbecame international standard. The key aspect of MPEG-7 is that it provides aframework of standardized tools that can be used to describe and efficientlymanage multimedia content. The standard specifies a set of descriptors anddescription schemes that can be exploited in applications such as video indexingand retrieval, content summarization, real-time content delivery, personalizedservices etc. Besides support for meta-data and text descriptions of audiovisualcontent, much focus in the development on MPEG-7 has been in the definition ofefficient content-based description and retrieval specifications. The description ofaudio-visual content includes low-level features that describe the signalcharacteristics, like the color, texture, shape and motion trajectories of objects in animage or a video, as well as the high-level features that describe the semantics ofthe content. The descriptions represent the content in a way that a search enginecan use these descriptions to find the material of interest to the user in a reasonabletime. Although feature extraction and search engines are required for completeapplications, these will be non-normative parts of the standard. The normative partof the standard will focus on the descriptions and their encoding. This will enabledevelopment of applications that will utilize the MPEG-7 descriptions withoutspecific ties to a single content provider or search engine.In this paper, we first study the MPEG-7 standard. We introduce the purpose,the object and the content of the MPEG-7 standard. We discuss the application ofthe MPEG-7 standard in the image retrieval technique and the descriptions of theimage visual features and the range of the application of the MPEG-7 standard.The compact color descriptor is one of color descriptors in the MPEG-7standard. We present some based-block compact color descriptors. The descriptorhas been evaluated for various applications such as image matching, browsing,video segment matching and sketch based search etc. Content-based search andretrieval techniques for multimedia content utilize various descriptors for differentimage features. The color feature, being one of the easily computable features isoften used in content-based search and filtering applications. But for manyapplications, a more compact descriptor that expresses the qualitativecharacteristics is sufficient. So we present a compact color descriptor based on thecolor histogram. This descriptor is obtained by applying the Haar wavelettransformation to a color histogram and binary quantizing the resulting coefficientsof the transform. The resulting descriptor is 63 bits long and is therefore verycompact. Objective measures developed by the MPEG-7 group to evaluate variousrepresentations sow that this descriptor offers a superior performance whencompared to other descriptors of comparable size.The extraction process of the compact color descriptor consists of four stages.The first stage is to divide the image into many of 2×2 or 4×4 sub-blocks, we canobtain the sum of each sub-block, then the value that the sum divides by 4 isrepresented this sub-block. The second stage is to obtain a color histogram using asuitable color space and quantization scheme. The histogram can be a one, two orthree dimensional histogram. The third stage is the computation of the Haartransform coefficients of the histogram. This stage does not depend on the nature ofthe color space or the quantization table used in the first stage. The Haar transformcoefficients are obtained by taking the inner product of the basis function with thegiven histogram. The fourth stage is the binary quantization of the resultingcoefficients.We use the sub-blocks to deal with the image and utilize the RGB color space.Based on the study of the each color, we have four quantification centers(32,96,160,224). Then we get a 64-bin histogram, which is transformed byhierarchical Haar wavelets. The Haar wavelets are computationally very simple toimplement, because the computation does not involve any multiplication.Field Programmable Gate Arrays ( FPGA ) are common devices forreconfigurable computing. They permit to satisfy, demanding applications thatnecessitate combining speed with flexibility. FPGA is a representative device thatadapts the development of the digital times.It is widely used in the digital systems.In fact, flexible architectures are very important for the news multimediaapplications. In the rest part of the paper, we use the FPGA to design hardwarestructures of the compact color descriptor. For adapting the implementation, weneed to improve the arithmetic. The implementation architecture is composed onthree blocks: quantification, histogram computation and Haar transformation,which are discussed in detail and designed by modules. We give some solutionsthat are synthesized and simulated.The compact colour descriptor can be used for the based-content imageretrieval system. Because of the binary value, so it only needs xor computation andcounts the number of the '1' when matching the values. This computation is verysimple, and suits fast retrieval system.By studying the arithmetic of the MPEG-7 compact color descriptor anddiscussing deeply the design technique of the complex system-on-chipimplementation, this paper gives the important reference value for content-basedmultimedia retrieval and the hardware implementation of MPEG-7 descriptors.Meanwhile it is very important to design large scale system-on-chip in the future.