SU-G-IeP2-01: A Method to Reduce Information Loss in JPEG Compressed Medical Images Using Richardson-Lucy Deconvolution

Abstract

Purpose: To minimize the information loss of medical images compressed by JPEG using Richardson-Lucy (RL) deconvolution. Methods: We first generated reference images having three different noise structures; low-passed noise, high-passed noise, and white noise, which mimic noise structures for different medical imaging modalities. Each reference image was filtered by a Gaussian function and compressed/decompressed by JPEG 2000 algorithm. After decompression, RL deconvolution was applied to recover the reference image using the same Gaussian function. The goal of the RL deconvolution is to restore the reference image from the decompressed image as similar as possible. For each reference image, the optimal kernel size of the Gaussian function and iteration number of the RL deconvolution were found by brute-force searching. The optimal parameters were chosen when the Structural Similarity Index Metric (SSIM) value between reference image and restored image by the proposed method was maximized. We also compared the SSIM value of decompressed conventional JPEG image. Results: For each reference image, proposed algorithm produces a higher SSIM value than conventional JPEG image, demonstrating the information loss by JPEG can be reduced by a simple deconvolution technique. Conclusion: In this work, we present a simple method to prevent the information loss caused by JPEG algorithm in medical images. We only tested our method using a simple Gaussian function, but using different functions optimized for different noise structures would improve the performance of the proposed method, which will be a subject of our future research. This research was supported by the MSIP(Ministry of Science, ICT and Future Planning), Korea, under the IT Consilience Creative Programs (IITP-2015-R0346-15-1008) supervised by the IITP (Institute for Information & Communications Technology Promotion), Basic Science Research Program through the National Research Foundation of Korea (NRF) unded by the MSIP(2015R1C1A1A01052268) and framework of international cooperation program managed by NRF(NRF- 2015K2A1A2067635).