GPU implementation of biquadratic B-spline image resizing
Shen Li · Journal of Mechanical & Electrical Engineering · 2010
The classical way of obtaining a resized image is by using kernels,such as image replication,bilinear interpolation and bicubic interpolation.Image replication is easily implemented but with blocking artifacts when resized,and bilinear interpolation can cause the blurring of sharp edges due to the lowpass filter effects,while bicubic interpolation has better visual effects but with more calculation and inefficiency.Aiming at the inefficiency of bicubic interpolation,an algorithm of resizing an image via biquadratic B-spline was proposed,which was based on the interpolation on the local area in the source image and avoids the phase distortion phenomenon by using 4 sample points to distinguish with conventional biquadratic interpolations.Due to the locality of the algorithm,it can be implemented in parallel,which guarantees the high efficiency of this approach.The results indicate that the new algorithm can almost has the same effects compared with the bicubic interpolation and can achieve about 10 times improvement in efficiency.