Visually Lossless Compression of Retina Images

Digital images of a rather high resolution are widely used in modern medical practice. Due to their large size, there exists necessity to compress them before storage or transmission via communication lines in telemedicine. Possibilities of lossless compression are limited and one often has to apply lossy compression with providing acceptable diagnostic quality of compressed data (with ensuring visually lossless compression). This paper proposes ways to carry out such a compression in one iteration, i.e. quickly enough with application to retina images. An efficient coder based on discrete cosine transform (DCT) in 32x32 pixels blocks is analyzed. It is shown that mean squared error (MSE, or PSNR (peak signal to noise ratio) of introduced distortions can be predicted by estimating distribution of alternating current (AC) DCT coefficients in a limited number of 8x8 pixel blocks and very fast processing of these DCT coefficients. We present approximating (predicting) curves obtained by regression of several types of simple functions into scatter-plots. This allows setting coder parameter (quantization step - QS) to provide a desired MSE. Applicability of the proposed way of prediction approach is demonstrated experimentally for real-life retina images.