Segmentation based image compression of brain magnetic resonance images using visual saliency

Abstract

Listen

Translate article

The coding of medical images is of paramount importance due to an exponential increase in digital imaging data. The realization of these coding methods is highly challenging as the key fact is to maintain the visual quality by preserving the clinically critical information and also reducing the storage space at the same time. To address these issues, the current paper proposes an efficient non-uniform compression algorithm based on visual saliency that emulates the human visual system. This hybrid technique works in two phases. In the first phase, an automatic saliency-based Fuzzy C-Means clustering algorithm (sal_FCM) is designed for Region of Interest (ROI) detection and extraction. In the second phase, the encoding of ROI and the background are carried out using the SPIHT algorithm (ROI-SPIHT) at a high and low bit rate respectively. To curtail the computational complexity of the proposed ROI-SPIHT algorithm, a wavelet-based on a Lifting scheme is used. The empirical evaluation was carried out on the BRATS dataset where the results suggest that the proposed approach accurately identifies the ROI with comparatively better visual quality and compression ratio.