Multistaged gradient based scaling technique

Digital zooming is a scaling operation that is performed for better view of the image. Scaling an image for studying its features is a general practice in almost any field that practices image oriented study for analysis and conclusions. When an image is scaled to study it, it is expected that the underlying scaling algorithm has to preserve the features of the original image. Algorithms like Nearest Neighbor (pixel replication) make the image too much blocky. Most commonly used image scaling algorithms (like bilinear), when observed, make the scaled image look blurred i.e. they introduce the effects of aliasing on the sharp corners, when scaling operation is performed. If the algorithm (example: lancoz) is designed to preserve the edges (high frequency components) then the plain regions (low frequency components) adjacent to edges are disturbed. The scaling method presented in this paper is designed to render scaled images with least artifacts. The approach implements restricted smoothening and thus preserve the edges in the scaled image. The proposed approach achieves this using 3 staged gradient based interpolation techniques. From the experimental results, it is shown that the proposed scheme provides very low computational complexity as compared to the conventional scaling techniques like nearest neighbor, bilinear and bi-cubic. Specifically, with the scaling factor of 2 (X2) 82.75% complexity can be reduced as compared to bilinear and 96.77% can be reduced as compared to bi-cubic. The proposed scaling method is applied on several medical images. The resulting images demonstrate the algorithm´s ability to magnify an image while preserving edges.