Parallel Interpolation Of Edge-Directed Image Interpolation On A GPU

Interpolation is a very important method in image processing. It is widely used in Image Magnification Image Denoising and De-interlaced. There are many methods for interpolation, but traditional method will be lost detail information because it has characteristics of low pass filter. So the edge-directed methods receive more and more attention. The edge-directed interpolation scheme is a non-iterative, orientation-adaptive method to enhance image resolution with better visual effect than conventional interpolation methods. It interpolates the missing pixels based on the covariance of a high-resolution image estimated from the covariance of the low-resolution image. In spite of the impressive performance, the computational complexity of covariance-based adaptation is significantly higher than that of the conventional linear interpolation algorithms. In this paper, we propose a GPU-based massively parallel version of the edge-directed interpolation scheme, and our main work can be summarized follows:(1) In order to reduce the complexity of the edge-directed image interpolation, we propose a GPU-based massively parallel version of the interpolation method. Each thread computes one missing pixel in high-resolution image. We have a successful of parallel implementing of the edge-directed interpolation on GPU. And a speed of12.2x is achieved with shared memory.(2) Although, the method can parallel implement on CUDA device. But the speed-up is trapped. So we use the computing resource to optimizate the parallel scheduling scheme. First the register is instead of shared memory to store the intermediate variable. Then the asynchronous transmission is used to hide the time of I/O. Fortunately, the speed-up is up to61.7x with optimization.(3) The edge-directed image de-interlaced actually is a method of image interpolation. Image de-interlaced means use interpolation method to recover the whole image from the image missing the even number line or odd-numbered line. We use GPU-based massively parallel the de-interlaced and we received speed-up of51.5x.