Bio-inspired FPGA architecture for self-calibration of an image compression core based on wavelet transforms in embedded systems

Abstract

A generic bio-inspired adaptive architecture for image compression suitable to be implemented in embedded systems is presented. The architecture allows the system to be tuned during its calibration phase. An evolutionary algorithm is responsible of making the system evolve towards the required performance. A prototype has been implemented in a Xilinx Virtex-5 FPGA featuring an adaptive wavelet transform core directed at improving image compression for specific types of images.An Evolution Strategy has been chosen as the search algorithm and its typical genetic operators adapted to allow for a hardware friendly implementation. HW/SW partitioning issues are also considered after a high level description of the algorithm is profiled which validates the proposed resource allocation in the device fabric.To check the robustness of the system and its adaptation capabilities, different types of images have been selected as validation patterns. A direct application of such a system is its deployment in an unknown environment during design time, letting the calibration phase adjust the system parameters so that it performs efcient image compression. Also, this prototype implementation may serve as an accelerator for the automatic design of evolved transform coefficients which are later on synthesized and implemented in a non-adaptive system in the final implementation device, whether it is a HW or SW based computing device.The architecture has been built in a modular way so that it can be easily extended to adapt other types of image processing cores. Details on this pluggable component point of view are also given in the paper.

The DLMT hardware implementation. A comparative study with the DCT and the DWT

“…However, compressed image have been still scaled (here, scale has the same meaning than DWT level or decimation process) by 2 or even by 4, in order to get 128x128pels or 64x64pels images respectively and to be able to compare these results although there is not a noticeable improvement. In those cases the PSNR decreases even to 22,50dB which are yet comparable to the results obtained in [13] and [14]. In all those experiments a very simple interpolation algorithm has been employed for recovering the original image size, based on the duplication of the neighbor pel.…”

“…The prototype platform used in [13] and [14] however is an ML507 development board, which contains a Xilinx Virtex 5 (XCVFX70T) FPGA device with an embedded PowerPC processor, responsible of running the Evolutionary Algorithm described. In that implementation, one result is obtained each two cycles, so the computation time of the whole wavelet at 150MHz clock cycle is 0,873msec.…”

“…Besides, the DLMT does not need to be adapted to the image type as reported in [13] and [14]. In the same way, the DLMT improves the hardware cost because a medium size low-cost device can be used (e.g.…”

“…All the results can be compared to the Evolutionary Algorithm (EA) proposed in [13] and [14]. Where the best fit PSNR for a set of 80 test images is 36,1490dB and an average fit PSNR of 29,8079dB after 500 mutations is obtained.…”

“…Furthermore, taking into account the calculation algorithms which have been proposed in [11], based on the use of multiple stages of signal filtering, some blurring effects and ringing at the edges of the image would arise. The lifting scheme proposed by Sweldends [12], and used in [13] and in [14] reduces the computational cost of the transform but the need of adapting the calculation algorithms to the type of signal is still necessary.…”