Comments on and corrections to unified VLSI architecture for photo core transform used in JPEG XR’

Abstract— In the aforementioned Letter [ 1], Zhang et al. proposed a very large-scale integration architecture to implement the three elementary 2 × 2 transform operations for the photo core transform used in JPEG extended range (XR), which are 2D 2 × 2 Hadamard transform, 1D rotate (T_odd), and 2D rotate (T_odd_odd). Pseudocodes for functions T_odd and T_odd_odd from JPEG XR standard [ 2] are presented in Listings 1 and 2 (Fig. 1), respectively. Since subtractions appear in all the computation steps, the negative sign ‘–’ needs to be added to appropriate data flows of the diagrams in Figures 1 and 2 in [ 1]. In addition, in the last paragraph of page 1, the authors claimed that the module in the dotted box in Figure 1 can be realised using 16 adders and 4 multiplexers. It should be 16 adders and 7 multiplexers. Furthermore, since T_odd and T_odd_odd in JPEG XR require constant additions as shown in Listings 1 and 2, a con-stant vector needs to be added to for T_odd and T_odd_odd transforms, respectively. This can be done using adders and shifters in a subsequent phase and does not affect the hardware architecture of the method. < final year projects >