Research On Image Interpolation Based On Gradient Information

In recent years,with the rapid development of computer vision,computational visualization,artificial intelligence and virtual reality,image interpolation has been widely used in high-definition television technology,smart home technology,digital projection technology and image processing software.In the image,the edge structure can convey most of the effective information described by the image,while the gradient vector can be used to accurately describe the edge structure of the image.Therefore,it becomes a new hotspot to exploit image gradient information to achieve image interpolation,which provides a new way of thinking to solve the tough problems in the field of image interpolation.The traditional image interpolation algorithms are widely used in many fields because of the simple mathematical model and the calculation.However,there are some drawbacks in the calculation of precision and interpolation effects.For example,it is likely to produce blurred details and jagged edges.As the basic part of image processing,the quality of image interpolation is of great significance to the subsequent operation of image processing.To a great extent,the quality of the edge has a direct effect on the display quality and visual effect of the interpolated image.And the precision of the gradient value also plays an important role in the description of the image edge.Therefore,this thesis obtains gradient of the low-resolution image by constructing the bi-cubic fitting surface,which effectively improves the precision of the gradient.To get the interpolated image with higher precision,the linear surface is constructed with the gradient weighting constraint,according to the relationship between the gradient and the intensity of the image.Considering the characteristics of the image that the intensity changes quickly along the gradient direction,this thesis proposes an image interpolation algorithm,where the interpolated image is calculated via the linear surface weighted by gradient.First,the proposed algorithm constructs the bi-cubic surface to calculate the gradient of the low-resolution image,in order to improve the precision of the image gradient.And the gradient of the image is obtained by the bicubic interpolation.Then,the algorithm adjusts the gradient of the high-resolution image locally,which fully considers the spatial correlation between the gradient distribution and the pixel distribution,making the gradient distribution of the interpolation image become reasonable.Finally,the pixels to be estimated in the high-resolution image is divided into three categories.The first class can be directly assigned by the corresponding pixel in the low-resolution image.And,the other are calculated via constructing the linear surface which is weighted by neighbor pixels.The weighted value of the pixel is determined by its spatial distance in the gradient direction of the estimated pixel.The smaller the spatial distance is,the greater the weight.In the experiment,this thesis has tested different kinds of images,including synthesis images,medical images,and natural images.The experimental results show that the algorithm has obvious advantages compared with the representative interpolation algorithms in recent years,both in visual effort and data analysis.This algorithm can maintain clear edge structure and texture details,significantly enhance the overall quality of the interpolation image.