Unsupervised regions of interest extraction for color image compression

Abstract

A novel unsupervised approach for regions of interest (ROI) extraction that combines the modified visual attention model and clustering analysis method is proposed. Then the non-uniform color image compression algorithm is followed to compress ROI and other regions with different compression ratios through the JPEG image compression algorithm. The reconstruction algorithm of the compressed image is similar to that of the JPEG algorithm. Experimental results show that the proposed method has better performance in terms of compression ratio and fidelity when comparing with other traditional approaches.OCIS codes: 100.0100, 100.2000100. , 150.0150, 150.1135 At present, lossy compression of still images is one of the research focuses in the digital image-processing field. Under the condition of the limited storage capacity of the imaging system and the constrained information transmission wide band, traditional uniform compression ratio (CR) algorithms are unable to consider both the requirements of high CR and image reconstruction quality. The non-uniform image compression methods based on regions of interest (ROI) apply lossless or high-fidelity compression to ROI while adopting a high-CR method to the background, which results in the effective compression of image redundancies, with important information being preserved as much as possible. Discriminating the ROI from the background is important. The most direct method to achieve this is to allow the user to select the interesting regions [1] . However, such an image-processing system may become quite complicated. Unsupervised extraction of a specific ROI from an image is an important procedure for computer vision and image-processing algorithms.Human beings have the remarkable ability to determine ROI in complex scenes quickly. This ability is called "selective visual attention mechanism" (SVAM). Therefore, introducing SVAM in unsupervised ROI extraction is important and necessary because it can reduce computational complexity [2] , save computational resources, and improve the efficiency of image processing.In this letter, a novel, non-uniform image compression approach based on an unsupervised ROI extraction algorithm is proposed. The algorithm can successfully distinguish ROI from the original image. When ROI have been extracted, the ROI and other regions are encoded with different CRs[3] using a popular image compression algorithm.The proposed approach adopts the intersection of the saliency map produced by a modified Itti-Koch visual attention model and the segmentation result of the clustering analysis algorithm to determine ROI. A flowchart of the proposed model is shown in Fig. 1.If peaks in the saliency map overlap with regions determined by the image segmentation, the ROI is extracted based on these regions. One of the most important advantages of the proposed model is that its execution is entirely unsupervised.The original Itti-Koch model of visual attention uses the dyadic Gaussian pyramid to subsample the input image [4] . In contrast, in the propo...