| Image scaling, also known as image resampling, is one of the key technologies in digital image processing. There are two kinds of digital image scaling algorithms including pixel-based algorithms and edge-based algorithms. The former algorithms easily result in edge blurring in target images due to the low-pass filtering effect, thus the latter ones attract more attention as human eyes are more sensitive to image edges according to the HVS study. In spite of better effects in edges and details, edge-based algorithms are of high complexity, making them difficult for hardware implementation and real-time image scaling. Besides, arbitrary expansion can't be realized since most of them can only achieve 2k zooming.In order to protect characteristics of image edges as well as obtaining arbitrary expansion, this paper presents an adaptive image scaling algorithm based on the cubic Lagrange interpolation, where variances are introduced to measure the correlations among original pixels. The smaller the variance is, the greater the correlation becomes, and vice versa. In the proposed algorithm, interpolations are carried out along the line direction and the field direction, respectively. In each interpolation, the variances of the three original sets of pixels around the target pixel are firstly computed, and then the set with the minimum variance is selected to calculate the gray value of the target pixel using third-order Lagrange interpolation formula. In hardware implementation, quadratic image scaling kernel structure is employed, and the image scaling engine and Farrow structure based Lagrange interpolation kernel are further presented, all of which reduce the hardware complexity significantly.The experimental results show that the proposed algorithm not only leads to higher PSNR and more vivid edges of target images, but also decreases the complexity greatly, which makes it suitable for hardware implementation and real-time image scaling. |
