Image interpolation using support vector machines

Image interpolation has a wide range of applications in remote sense, medical diagnoses, multimedia communication, and other image processing fields. Support vector machines (SVMs) have been used successfully for various supervised classification tasks, regression tasks, and novelty detection tasks. In this chapter, support vector machines based image interpolation schemes for image zooming and color filter array interpolation are discussed. Firstly, a local spatial properties based image interpolation scheme using SVMs is introduced. After the proper neighbor pixels region is selected, SVMs are trained with local spatial properties that include the mean and the variations of gray value of the neighbor pixels in the selected region. The support vector regression machines are employed to estimate the gray value of unknown pixels with the neighbor pixels and local spatial properties information. Some interpolation experiments show that the proposed scheme is superior to the linear, cubic, neural network and other SVMs based interpolation approaches. Secondly, a SVMs based color filter array interpolation scheme is proposed to effectively reduce color artifacts and blurring of the CFA interpolation. Support vector regression (SVR) is used to estimate the color difference between two color channels with applying spectral correlation of the R, G, B channels. The neighbor training sample models are selected on the color difference plane with considering spatial correlation, and the unknown color difference between two color channels is estimated by the trained SVR to get the missing color value at each pixel. Simulation studies indicate that the proposed scheme produces visually pleasing full-color images and obtains higher PSNR results than other conventional CFA interpolation algorithms.