Image Zooming using Corner Matching on Grayscale and Color Images

Originally inspired by the imaging needs of UND's Open Prototype for Educational Nanosats (OPEN) satellite program. The goal of this research is to direct the choice of an image interpolation/zoom algorithm. The image is scanned and overlapping 3x3 blocks of pixels are analyzed looking fir "L" patterns. Such a pattern indicates that the value of the center pixel be changed transforming the "L" pattern into a triangle pattern. We compare this approach against different types of single-frame image interpolation algorithms, such as zero order-hold (ZOH), bilinear, bicubic, directional cubic convolution interpolation (DCCI) approach, and a neural network approach. We use the peak signal-to-noise ratio (PSNR), mean squared error (MSE), Structural Similarity Index (SSIM), and Feature Similarity Index (FSIM) as the primary means of comparison. Tests cases include gray-scale and RGB images. In all test cases the proposed method resulted in a lower MSE, higher PSNR, and higher SSIM than the other zoom methods applied. The FSIM returned a higher value in every case for the ZOH method, but this was only when FSIM was calculated to the fourth decimal. Overall, this method results in a more accurate image after zooming than the other methods.