The compression of images helps in reducing transmission time and storage capacity costs. The JPEG 2000 standard is based on wavelet transforms, and offers several advantages over the previous JPEG standard. Notably, it supports region of interest (ROI) coding, where the pixels belonging to the ROI are encoded with better quality than the rest of the image. Maxshift and general scaling are two of the methods proposed to encode the ROIs. However, they do not permit selection of the relative importance of ROI and background. Further, the different ROIs cannot be encoded at different quality levels. This paper proposes a technique, selective coefficient mask shift, to overcome these drawbacks.
The scheme is complaint with JPEG 2000, and utilizes the properties of the wavelet coefficients at various decomposition levels to achieve variable quality encoding for ROIs. It shifts the coefficients associated with the ROI relative to each sub-band, according to its priority. This permits reconstruction of the image where different ROIs appear at different levels of quality. However, the scheme introduces an additional constraint that the number of decomposition levels should be greater than or equal to the number of ROIs. This constraint somewhat increases the complexity of encoding when a large number of ROIs are selected.
The performance of the scheme is evaluated by choosing four ROI masks. It applies a background reconstruction algorithm to remove the artifacts introduced. The results for the “Barbara” image and computed tomography (CT) images show that different ROIs can be encoded with different quality levels.