A Machine Learning Algorithm for Image Compression with application to Big Data Architecture: A Comparative Study

Abstract Background and Objective: In this paper, our paradigm is based on the workflow proposed by Tchagna et al and we now propose a new compression scheme to implement in this step of the workflow. The goal of this study is to propose a new compression scheme, based on machine learning, for vector quantization and orthogonal transformation, and to propose a method to implement this compression for big data architectures. Methods: We propose developed a machine learning algorithm for the implementation of the compression step. The algorithm was developed in MATLAB. The proposed compression scheme performed in three main steps. First, an orthogonal transform (Walsh or Chebyshev) was applied to image blocks in order to reduce the range of image intensities. Second, a machine learning algorithm based on K-Means clustering and splitting method is utilized to cluster image pixels. Third, the cluster IDs for each pixel is encoded using Huffman coding. The different parameters used to evaluate the efficiency of the proposed algorithm are presented. We used Spark architecture with the compression algorithm for simultaneous image compression. We compared our obtained results with literature results. The comparison was based on three parameters, Peak Signal to Noise Ratio (PSNR), MSSIM and computation time. Results: We applied our compression scheme to medical imagery. The obtained results of parameters for the Mean Structural Similarity (MSSIM) Index and Compression Ratio (CR) averaged 0.975% and 99.75%, respectively, for a codebook size equal to 256 for all test images. In comparison, with codebook size equal to 256 or 512, our compression method outperforms the literature methods. The comparison suggests the efficiency of our compression method. Conclusion: The MSSIM showed that the compression and decompression operation performed without loss of information. The comparison demonstrates the effectiveness of the proposed scheme

Author(s): Kouanou AT,Tchiotsop D,Tchinda R and Tansaa ZD

Abstract | Full-Text | PDF

Share This Article