Research And Design Of Image Scaling Algorithm For Interested Objects

The development of network and communication technology has brought portability of mobile devices and video capture devices, and a higher request on high-quality images and video displaying at a variety of mobile terminals. People want to keep the source image information after scaling, and to get good visual effect.For currently existing image scaling, this paper presents a sensitivity object recognition and triangular mesh deformation scaling methods. The image objects include buildings, trees and cars. The paper extracts color, texture, shape and special features of training samples after SVM training, forms sensitive object features as the basis for image recognition. When read a test image, it recognizes the object, considering the size and position of objects in the image, then calculates the sensitivity values of the objects.In image scaling, the paper proposes a triangulat ion-based image scaling algorithm with constrain loss function. The existing image distortion scaling method in achieving grid deformation only take into account the shape of the grid while ignoring the distorted objects of the image. Firstly, it detects corners, and then uses Delaunay to get the initial triangular grid, and introduces a loss function of the mesh deformation to the grid deformation process. With this function, the scaling results can maintain good visual effects.Unlike conventional significant importance detection or identification, the paper calculates sensitivity values for objects of interest in image to achieve more accurate expression, and combines scaling loss function and triangular mesh deformation in image scaling, and achieves better visual effect.