Abstract: Since current image resizing algorithms via correlation detection do not fully consider the correlation directions and only adopt the constant correlation along the detected image edge,they cannot always reveal the potential correlations among the neighboring pixels.An algorithm of non-linearly zooming in an image is proposed,which is based on the correlation detection.The potential correlation directions are fully considered;and constant,linear and quadric correlations along the image edges are detected,which can describe local image edge features better.Experimental results show that the proposed algorithm can alleviate the image blur and accomplish high quality zooming-in effects.Furthermore it is a local algorithm and can be implemented in parallel on GPU.