Enhancing the Intra-prediction in JPEG-XR by Using Edge Information

The new image coding standard, JPEG-XR, uses an inter-block prediction mode to remove redundancy in the quantized transform coefficients. However, it is arguable whether the prediction modes in JPEG-XR are efficient enough for accurate prediction. This paper proposes a new inter-block prediction method which adapts to local image feature, i.e. edges. Since edges represent distinct image structures, they correspond to the local directional features within blocks. Taking advantage of this property, we propose a two steps inter-block prediction method. First, edges orientations are obtained. Then, these orientations are used to direct the inter- block prediction. The proposed prediction method is compared with that in JPEG-XR. The proposed method can reduce the encoding time and improve the PSNR.