Local Spatial Property Based Support Vector Machines Image Interpolation Scheme

Support Vector Machines (SVMs) have been engaged on image interpolation tasks recently. These methods employed only local pixel coordinates or neighbor pixels gray values as input properties and obtained poor quality result images. A novel SVMs interpolation scheme was proposed with increasing the local spatial properties as SVMs input information. At first a proper neighbor pixels model was selected. Then SVMs were trained with local spatial properties that include the average of neighbor pixels gray values and orientation variations between neighbor pixels. Finally the support vector regression machines estimated the gray value of an unknown pixel with the neighbor pixels and local spatial information. Some interpolation experiments demonstrated the effectiveness of the scheme.