Content perception binocular image zooming method based on grid deformation

Abstract

The invention discloses a content perception binocular image zooming method based on grid deformation. The method comprises the following steps: A, inputting an original left-eye view and an original right-eye view, and setting the resolution of an object image; B, respectively constructing uniform quadrilateral grids covering the image for the left-eye view and the right-eye view; C, respectively calculating grid-grade importance degree graphs of the left-eye view and the right-eye view; D, constructing a grid-grade linear characteristic set of the left-eye view and the right-eye view; E, according to the grid-grade importance degree graphs, establishing grid zooming energy functions; F, according to the grid-grade linear characteristic set, establishing linear characteristic constraints, and carrying out grid zooming energy function minimizing operation by use of least squares to obtain a left-eye view object grid and a right-eye view object grid after zooming; and G, mapping the original left-eye view and the original right-eye view respectively to an object grid by use of a mapping method, and outputting a final left-eye view and a final right-eye view. According to the invention, the linear characteristic of the zoomed image can reserved, important objects are prominent, and stereo information of a binocular image is maintained.

Description

technical field [0001] The invention belongs to the field of image processing, and in particular relates to a content-aware binocular image scaling method based on grid deformation, which can be applied to multi-scale reduction and enlargement processing of binocular images. Background technique [0002] With the promotion of 3D display devices of different specifications, in order to adapt to the playback requirements of display screens with different aspect ratios, the unequal scaling of binocular images has attracted the attention of scholars at home and abroad. Traditional content-aware image scaling methods focus on the scaling of 2D images. The research points include the accurate detection of salient areas, the avoidance of image distortion and fracture, the maintenance of main features and smoothness of the image, etc., which can be roughly divided into discrete and continuous scaling methods. The discrete methods include methods based on seam carving, block process...

AI Technical Summary

Problems solved by technology

However, after being processed by this method, straight-line objects in the image tend to be bent, which brings uncomfortable viewing experience to users

At the same time, the sparse correspondence between size-invariant feature conversion depends on the scene structure of the image, and this method has limitations in application scenarios.

[0004] In summary, the existing content-aware binocular image scaling methods obviously have limitations and defects in actual use, so it is necessary to improve

Images