Projection-based eigenvector decomposition for reduction of blocking artifacts of DCT coded image

We present a projection on convex sets (POCS)-based method for removing blocking artifacts in reconstructed DCT-encoded images at low-bit rate. Our method can be realized with reduced complexity by cancelling out IDCT and DCT pairs in the POCS iterations. Our key point is converting pixel-smoothness constraints across block-boundaries directly onto certain requirements for the DCT coefficients. This leads to new closed convex sets defined on an eigenvector decomposition of the projection-based matrix in the DCT domain. Experimental results show that our method achieves an enhanced decoding, both objectively and subjectively