Evolutionary Approach to Improve Wavelet Transforms for Image Compression in Embedded Systems

Abstract

A bioinspired, evolutionary algorithm for optimizing wavelet transforms oriented to improve image compression in embedded systems is proposed, modelled, and validated here. A simplified version of an Evolution Strategy, using fixed point arithmetic and a hardware-friendly mutation operator, has been chosen as the search algorithm. Several cutdowns on the computing requirements have been done to the original algorithm, adapting it for an FPGA implementation. The work presented in this paper describes the algorithm as well as the test strategy developed to validate it, showing several results in the effort to find a suitable set of parameters that assure the success in the evolutionary search. The results show how high-quality transforms are evolved from scratch with limited precision arithmetic and a simplified algorithm. Since the intended deployment platform is an FPGA, HW/SW partitioning issues are also considered as well as code profiling accomplished to validate the proposal, showing some preliminary results of the proposed hardware architecture.

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

“…The final time measured increases to 12 minutes, which means that the system is able to adapt to a new type of input data in this time. This may look like a huge amount of time, but as shown in our previous work, 7 between generations 100 and 300 the system has already evolved a better solution. This reduces the effective evolution time to between just 1.2 and 3.6 minutes.…”

“…This issue makes their implementation as a hardware embedded system highly unfeasible, which is precisely what this work addressed in a previous stage of the research, 7 to find an adequately tuned EA able to keep up with the quality of the transforms evolved in the State of the Art, but feasible enough to be implemented in an FPGA. The cut-downs done to the algorithm (a PC-based SW simulator version of the work reported in this paper) did undoubtedly affect the search performance, but a trade-off was found which validated the proposal, reporting 1.57 dB improvement over the D9/7.…”

“…Therefore, instead of dealing with the mathematical construction of new wavelets, an EA can be used to design/optimize the constituent wavelet filters. Our previous work 7 features a deep review of the current State of the Art. However, some brief comments are included here for the readers' convenience.…”

“…Table 1 shows the partitioning resulting from applying this design philosophy and the results obtained after doing a software simulation for 500 generations, which was addressed in a previous stage of this work. 7 Each of the EA operators (as will be introduced in Section 4) are shown together with further actions to be accomplished: recombination (of the selected parents); mutation (of the recombinant individuals to build up a new offspring population); evaluation (of each offspring individual); and selection (of the new parent population for the next generation).…”

“…This proposal and the thorough set of tests performed to tune the ES and find a suitable set of parameters for evolution to succeed, in the case of adapting the DWT to fingerprint images, can be found in our previous work. 7 The summary of these proposals and the resulting ES are reproduced here for the shake of clarity, pointing to the particular conference papers where they were first presented.…”