Efficient medical image compression/reconstruction applying the discrete wavelet transform on texturally clustered regions

Abstract

This paper suggests a novel image compression scheme, using the discrete wavelet transformation (DWT) and the k-means clustering technique, suitable for medical images. Moreover it suggests a novel reconstruction scheme based on Bayesian formalism. The goal is to achieve higher compression rates by applying different compression thresholds for the wavelet coefficients of each Detail DWT hand, in terms of how they are clustered according to their absolute values. This methodology is compared to another one based on preserving texturally important image characteristics, by dividing images into regions of textural significance, employing textural descriptors as criteria and clustering methodologies. These descriptors include cooccurrence matrices based measures. While rival image compression methodologies utilizing the DWT apply it to the whole original image, the herein presented novel approaches involve a more sophisticated scheme. That is, different compression ratios are applied to the wavelet coefficients belonging in the different regions of interest, in which either each wavelet domain band of the transformed image or the image itself is clustered, respectively. Regarding the first method, its reconstruction process involves using the inverse DWT on the remaining wavelet coefficients. Concerning the second method, its reconstruction process involves linear combination of the reconstructed regions of interest. Moreover, another more efficient variant of this second approach is proposed, which reduces blocking effects and is based on Bayesian formalism. An experimental study is conducted to qualitatively assessing all approaches in comparison with the original DWT compression technique, when applied to a set of medical images acquired from endoscopic video sequences.