Research and Hardware Design of MPEG-7 Texture Descriptor |
|
Author | DengHao |
Tutor | FuPing |
School | Jilin University |
Course | Communication and Information System |
Keywords | MPEG-7 Texture descriptior Co-occurrence matrix FPGA |
CLC | TP391.41 |
Type | Master's thesis |
Year | 2007 |
Downloads | 79 |
Quotes | 1 |
With recent advances in network, digital TV and multimedia technologies, more and more video data are being captured, produced and stored. Video information has the highest proportion, the most amount of information and the widest application prospect in multimedia information. Because video belongs to unstructured information and has the character of complex, hugeness, redundancy, it’s difficulty for people to find the needful video in great amount of information. so that a perfect retrieval method which can realize content based searching and fulfill demands of all kinds of multimedia customers becomes increasingly important in image search. The most current multimedia searching methods are by and large based on keywords or language description, which speed is very quick, but the amount of retrieval results would be very large, and it can’t complete the requirement of correctness. So many professional persons began to focus on the study of content-based multimedia re-trieval recently.A content-based multimedia retrieval system is introduced in this paper. This system is used for image retrieval of digital museum. For the feature extraction of images, a description tool is developed according to MPEG-7 standard. The system uses color, shape and texture information defined in MPEG-7 standard, as well as simple semantic annotations, to search images. And two search methods are used: Query-By-Example (QBE), that users must submit an image and the system return a group of matching images and Query-By-Keyword(QBK) that users can use simple language descriptions to retrieve images. The aim of the research of MPEG-7 is creating a kind of commonly description interface standard, which can satisfy the real-time (or not) requirement coming from various multimedia systems. The standard of MPEG-7 defined a great deal of data structure using to describe multimedia data. It can not only describe the low class characters (such as video and audio characters), but also it can describe the high-class characters (such as semantic, object and structure characters). In our papers research the standard of MPEG-7 and the standard descriptions, discuss the application of MPEG-7 standard in image retrieval technology, and the description method of image vision characters in MPEG-7 standard.Texture description is a bigness type in MPEG-7 description. The MPEG-7 standard has defined three types:“Texture Browsing Descriptor”,“Homogeneous Texture Descriptor”,“Edge Histogram Descriptor”. Base on MPEG-7 standard in our papers, we explicate the texture description which based on Co-occurrence matrix, in addition to the three texture description. Texture is an important character which is difficult to descript in image, usually, use statistics, structure, and spectrum to analysis. Gray level Co-occurrence matrix is a effectual Stat. method. It is a perfect method to descript distributing character in gray space in local image area, and descript space relativity between pixels position. Co-occurrence matrix is profit to hardware implementation in texture retrieval relative to wavelet transform and Fourier transform. So the texture description base on Co-occurrence matrix is a reasonable technology explore in MPEG-7 hardware implementation.The texture description retrieval base on gray level co-occurrence matrix consist of three steps: First, select appropriate quantization strategy in original images, gain the gray images after quantization. Second, calculate the gray images, gain the gray level co-occurrence matrix. Thirdly, calculate the feature gray level co-occurrence matrix, such as Angular Second Moment, Contrast, Maximum probability and so on.In our paper use uniform quantization to process the gray images, then calculate the gray level co-occurrence matrix. Using the gray level co-occurrence matrix character to calculate, gain the feature such as Angular Second Moment, Contrast, Maximum probability. Toward the feature such as Entropy, Sum of Squares: Variance, Correlation and so on, it is difficult to implement which involve division, evolution and square root.FPGA is an universal configurable calculate devices, it can satisfy the application situation which require both speed and flexibility. As the type of Programmable Logic Devices, It appeared increasingly suited to the digital development nowadays, and widely used in modern digital system design. In fact, to multimedia applications, the flexibility is very important. In the next part of our papers, we study the use of FPGA to achieve texture descriptors. To suit hardware implement, we must make the necessary improvements to the algorithm, put the descriptors extraction divided into three parts: quantification, the co-occurrence matrix calculation and characteristics calculation. We have a detailed discussion on every part, using modular design, discuss parts of the realization of the program, and do the synthesis and simulation.