Key Technology Research Of JPEG2000 Chip Implementation For Wireless Applications

As the latest international still image coding standard, JPEG2000 has a great performance on low-rate image coding, region of interest (ROI) coding, and progressive transmission, which are very suitable for wireless image transmission. However, the high complexity of JPEG2000 system restricts its excellent performance and widespread use. The two main features of wireless image transmission are the large-size image and the low bit rate, which are the starting point of the topic. In this topic, we research the key technology of JPEG2000 chip implementation by proposing several algorithms and designing high-speed circuits, significantly improving the performance of JPEG2000 system on coding large-size images on a low bit rate.First, the multi-input/multi-output circuits of 5/3 and 9/7 discrete wavelet transform (DWT) based on rearranging data are proposed. Based on the data-rearranging idea, the two-dimention (2D) 9/7 DWT circuit reads the image data by four lines in parallel, achieves the data match and parallel computation between several row and column processors, leading to the computing time of this circuit only slightly larger than that of reading data. The data-rearranging module has a simple and clear logic, making the circuit easily extended. The flipping structure is adopted to design the row and column processors, giving the circuit a shorter critical path and a lower distortion comparing to the lifting structure. Comparison with existed structures on work frequency, computation cycles and resource cost prove the high performance of the proposed structure. Based on this design, a 2D 5/3 DWT circuit with data-rearranging module is proposed, and it also obtains a high working frequency and a low memory cost.Second, the optimized background pretruncated method (OBPTM) and its circuit design are proposed to achieve JPEG2000 region of interest (ROI) coding. OBPTM introduces the reference parameter and the corrected one to accurately control the precision of background coefficients, obtaining high ROI quality. The calculation of the two parameters is given, and the corrected parameter is determined in terms of human vision system to improve the visual effect of reconstructed images. OBPTM also has a great performance on system computation, compliance with the standard, convenience for applications, and so on. To fill in the blanks of the hardware design of ROI coding, a circuit structure of OBPTM for rectangle ROI coding is proposed, and it can adaptively adjust the reference parameter by the application, bit rate and ROI area. The circuit design has a high throughput and can be easily embedded into the existed JPEG2000 hardware system.Third, the algorithm of embedded block coding with optimal truncation (EBCOT) is optimized by the OBPTM and its parallel coding structure is designed. With the analysis of the distribution characteristics of subband coefficients, we point out that the blank bitplanes of background codeblocks only contain certain information, and the number of blank bitplanes needed coding is also calculated. Then the EBCOT Tier-1 circuit skips the coding of the other blank bitplanes and significantly decreases the coding time of background codeblocks. The post-compression rate-distortion optimization (PCRD) circuit optimized by OBPTM is also proposed. On one hand, the calculation of rate-distortion slope for some truncated points is skipped, decreasing the calculation time of feasible truncated points. On the other hand, the searching time of the threshold value of rate-distortion slopes is decreased with the two-level searching method. At last, the pipeline structure of EBCOT is designed, implementing the parallel computation of PCRD and several EBCOT Tier-1 modules. As the largest amount of calculation of PCRD, the slope calculation process no more occupies the coding time of JPEG2000 circuitForth, the spatial filtering method is adopted to restrain the boundary artifact caused by the JPEG2000 multi-tile coding. The former filter is introduced to keep the original detail information of the tile boundary, and the latter filter is adopted to decrease the high-frequency components introduced by boundary extension of DWT. Compared to the traditional method that takes 1D array as the analysis object, the 2D matrix is adopted as the input model which is close to practice. Then, through the matrix expression of the former filter, DWT, quantization and the latter filter, and the establishment of a precise quantization distortion model, the statistical relationship between the input image and the mean square error of the reconstructed image is derived. Based on this statistical relationship, the curve of the distortion of boundary and non-boundary coefficients with the change of the former filter is obtained. At last, the optimal values of the former and latte filters are determined by proposing a new measure standard of the algorithm performance.Finally, a high-performance structure for JPEG2000 circuit is designed, and a subband coefficient buffer is introduced to implement the high-efficiency flexible connections of DWT, OBPTM and EBCOT circuits. The test results on FPGA show that the JPEG2000 circuit works stably on 150 MHz, and only cost 1.37 ms on coding a tile with size of 256×256. A system on chip for image coding and transmitting based on the JPEG2000 circuit is proposed, further verifying the performance of the coding circuit and the spatial filtering method.The adoptions of OBPTM and spatial filtering method significantly improve the image coding quality at a low bit rate. The high-speed hardware implementations of DWT, OBPTM, EBCOT optimized by OBPTM and JEPG2000 structure achieve a high coding speed. All in all, the JPEG2000 circuit designed in this topic can satisfy the requirements of wireless image transmission.