Research And Application Of The Autoregresstive Model-based Image Interpolation Algorithm

With the development of the computer technology and the greater demand on low-cost high definition picture, Image interpolation technology has been concerned more and more. Image interpolation is the process of turning low definition pictures into high definition ones under the model frame, which is used to recover the lost information in the picture. This technology has been applied extensively, including scientific research, industry, agriculture, biomedicine engineering, military, public security, judicature. culture and the arts etc, and it has solved many key problems in the area of image amplification, image de-nosing, super definition restructuring of image, etc.Having studied the conventional interpolation algorithms, this thesis proposes the concept of the autoregressive-based adaptive interpolation and the respective model. The basic ideas are:1)to construct the AR model based on the pixel structures of the image:to estimate the AR coefficients according to the relations among pixels from the low-resolution image, and assume the AR model derived from the low-resolution image can be adapted to the high-resolution image, based on the geometric duality between the low-resolution image and high-resolution image; to carry out calculations with the knowledge of the pixel values of the low-resolution image.2)to employ the idea of iteration; having got the initial interpolated values by applying the AR model, to consider the multiple constraints that the to-be-interpolated pixel is involved, respectively to construct multiple AR models, and to weightedly summarize their results; the process of iteration can be implemented many times, until the best result is achieved.3)in order to enhance the adaptation of the algorithm, the weighted least square method is adopted; to take the structures of the image into account, and to assign different weights to different pixels in the training window, based on their distance to center, thus the influence to the to-be-interpolated value by different pixels can be differentiated; the model to get the weight is studied, and the related parameters are also discussed. Simulation results demonstrate that this interpolation algorithm can generate interpolated images with excellent subjective quality as well as high PSNRs.