Abstract
This study addresses the problems dealing with medical images - to assist the doctors to collect/retrieve visually similar images for education usage, and to provide image communication solution for telemedicine.The quality of medical images is crucial in diagnosis. Unlike common natural images, medical images are more specific and could not be distorted during transmission and storage. Normally, medical images are taken under special condition and specific to the special part of human body. If we could utilize the domain knowledge of specific type of medical images, higher compression ratio could be achieved without losing any useful diagnostic information. In order to reduce the cost of storage and transmission bandwidth, we proposed a segmentation method to automatically locate the ROI area in a chest radiograph medical image, which is enclosed by ribs. By utilizing the knowledge of significant edge of ribs, our method can successfully locate the area that is bounded by ribs. The located ROI area is then compressed by the improved SAMAR method, and the non-ROI area is compressed by JPEG.To help radiologists to collect useful medical images for the tutorial of differential diagnosis, we proposed a content-based image retrieval system to cluster visually similar image patterns, and to mine the similarity and differentiation between similar syndromes. The system uses visual-based user interface to allow the user to enter or query an image by selecting the region of interest (ROI) regions; and uses a neural network method to classify the relationship between the images stored in database. The system will output a set of candidate images that are textural-similar to the query image. In the implementation, we extract the major 2-D FFT coefficients to represent the texture features and uses Kohonen self-organizing network to cluster those extracted FFT coefficients.In the field of telemedicine, transmitting medical video stream from remote hamlet to clinic is necessary and important. For some consequential operations, we need to build a failure-free communication during some emergency operations. We propose a scalable low bit rate failure-free video-coding algorithm for such transmission. For each video sequence in the failure-free video transmission, two channels are utilized to transmit the same video source. In the receiver end, the video can be reconstructed by any single encoded bitstream. While receiving two encoded bitstreams, the image quality can be improved by combining two bitstreams. In this study, we develop a motion vector refinement technique to improve the image quality.To meet the criteria of real-time transmission, we proposed a complexity-adaptive fast block-matching algorithm that allows users to terminate processing at any target computational complexity. Two complexity-adaptive implementations, frame level complexity allocation and block level complexity allocation, are developed to approach the global complexity-distortion optimization. In addition, a buffer control strategy was proposed to dynamically adjust the target complexity.