Prediction error based sharpening scheme for efficient 2-D up-sampling performance

This paper presents an efficient prediction error based sharpening scheme to enhance the high frequency regions of a low resolution image so as to effectively reduce the blurring in its corresponding high resolution counterpart. Most of the interpolation schemes available in the literature produce blurring artifacts in the up-sampled, high resolution image. In order to resolve this problem, the proposed scheme extracts the degraded high frequency information due to the up-sampling and then superimposes it on the original low resolution image to lessen the blurring in any subsequent up-sampling process. The proposed scheme extracts the degraded high frequency information by purposely down-sampling the given low resolution image and then restoring it back to its original size using Lanczos-3 interpolation. The lost high frequency information or the prediction error is obtained by subtracting the restored, reconstructed image from the original. The prediction error contains all the high frequency details that are lost during the up-sampling process. Hence, by adding the weighted version of the lost high frequency details with the original prior to any further up-sampling, the degree of blurring can be considerably reduced in the up-sampled, high resolution image. Experimental results show the proposed method gives better objective and subjective performance than other existing interpolation schemes.