Generalized bitplane-by-bitplane shift method for JPEG2000 ROI coding

One interesting feature of the new JPEG2000 image coding standard is support of region of interest (ROI) coding using the maximum shift (Maxshift) method, which allows for arbitrarily shaped ROI image compression without shape coding or explicitly transmitting any shape information to the decoder. The major disadvantage of the Maxshift method is that it cannot adjust the scaling value which determines the degree of relative importance between the ROI and the background wavelet coefficients. The bitplaneby-bitplane shift (BbBShift) method was introduced to support both arbitrary ROI shape and arbitrary scaling without shape coding. In this paper, we propose a generalize BbBShift (GBbBShift) method, which delivers much more flexibility than both Maxshift and BbBShift for “degree-of-interest” adjustment of the ROI with insignificant effect on coding efficiency and computational complexity. Experiments show that it can provide significantly better visual quality than Maxshift at low bit rates. GBbBShift is not compliant with the current JPEG2000 definitions. In order to use it, a new ROI coding mode would need to be added to the standard. 1. JPEG2000 ROI CODING Region of interest (ROI) image coding allows for encoding the ROIs in an image with better quality than the background (BG). Two kinds of ROI coding methods are defined in the new JPEG2000 image coding standard [1]–[4] — the general scaling based method and the maximum shift (Maxshift) method. In the general scaling based method, the wavelet transform is applied to the image at the encoder and the resulting coefficients not associated with the ROI are scaled down (shifted down) so that the ROI-associated bits are placed in higher bitplanes. During the embedded bitplane coding process, the bits in the higher bitplanes are placed before those in the lower bitplanes. The scaling value and the shape information of the ROIs are also added into the encoded bitstream. At the decoder, the bitplanes are reconstructed and the non-ROI associated coefficients are scaled up to their original bitplanes before the inverse wavelet transform is applied. If the encoded bitstream is truncated or the encoding/decoding process is terminated before the image is fully encoded/decoded, the ROIs will have a higher quality than the BG. The relative importance of the ROIs and the BG is determined by the scaling value s, which defines the number of bitplanes to be shifted. Fig. 1 shows how the bitplanes are shifted in the general scaling based method. There are three major drawbacks of the general scaling based method. First, it is not convenient to deal with different wavelet subbands in different ways, which is sometimes desired by the users. Second, it needs to encode and transmit the shape information of the ROIs. This significantly increases the complexity of encoder/decoder implementations. Third, if arbitrary ROI shapes are desired, then shape coding will consume a large number of bits, which significantly decreases the overall coding efficiency. The current standard attempts to avoid this problem and only defines rectangle and ellipse shaped ROIs [2], which can be coded with a small number of bits. However, this limits the application scope of ROI coding because in many real-world applications, ROIs are usually associated with certain objects in the image, which generally have arbitrary shapes. A very effective solution, the Maxshift method [1], [3]–[8], was proposed for JPEG2000, which does not require any shape coding or any shape information to be explicitly transmitted to the decoder. In Maxshift, the scaling value, s, must be chosen to satisfy s ≥ max(Mb) [1], where max(Mb) is the largest number of magnitude bitplanes for any coefficient. After scaling, all significant bits associated with the ROI will be in higher bitplanes than all the significant bits associated with the BG [1]. Fig. 2(a) demonstrates this method. At the decoder, the ROI/BG coefficients can be identified simply by looking at the coefficients’ magnitudes. All non-zero coefficients that are found to be lower than the sth bitplane are known to belong to the BG. The non-ROI coefficients are scaled up by s bitplanes before the inverse wavelet transform is applied. With Maxshift, it is also easy to treat different wavelet subbands differently. For example, the encoder can include entire low-frequency subbands in the ROI mask and encode a uniform low-resolution version of the image at an early stage of the encoded bitstream. The ROI/BG distinction is made only at high frequency subbands. The major limitation of the Maxshift method is that it does not have the flexibility to allow for an arbitrary scaling value to define the degree of relative importance between the ROI and the BG wavelet coefficients. This means that in all the subbands, where the ROI/BG distinction is applied, no information about the BG coefficients can be received until every detail of the ROI coefficients has been fully decoded, even if the detail is imperceptible random noise (which may happen in reversible coding mode or irreversible coding mode with very small quantization step size). 2. GENERALIZED BBBSHIFT 2.1. BbBShift Scheme In [9], we proposed a bitplane-by-bitplane shift (BbBShift) method. Instead of shifting the bitplanes all at once by the same scaling value s as in Maxshift, BbBShift shifts them on a bitplane-bybitplane basis. An illustration of the BbBShift method is shown in Fig. 2 (b). Two parameters, s1 and s2, are used in BbBShift. The sum of s1 and s2 must be equal to the largest number of magnitude