Discussions on Implementing Iterative Hard Thresholding Algorithm

Compressive sensing is a potential technology for lossy image compression. With a given quality, we may represent an image with a few significant coefficients in the transform domain. When the number of the significant coefficients is much less than the number of the pixels, the assumption of sparse representation is satisfied. Based on the sparse modeling theories, an image could be sensed with a relatively simple hardware and reconstructed with a powerful computer. We are interested in how to implement the iterative hard thresholding algorithm. The formula is not complex, but the implementation is not straightforward when the image resolution is high. Therefore, the computation complexity and the memory consumptions are analyzed. With the analysis result and the implementation experiences, we discuss the issues that should be considered carefully when implementing the algorithm in this paper.