Depth image super-resolution reconstruction method based on high-quality edge information guidance

Abstract

The invention relates to a depth image super-resolution reconstruction method based on high-quality edge information guidance. The method comprises the following steps that: preprocessing one group ofhigh-resolution example images to obtain high-resolution and low-resolution image block pairs; carrying out joint training on the high-resolution and low-resolution image blocks to construct the dictionary pair of the high-resolution and low-resolution image blocks; inputting a low-resolution test image, amplifying an interpolation to a target size, using impact filtering to process the interpolated image, and extracting the edge of the filtered image using a training dictionary to carry out sparse representation on the edge image to construct a high-quality edge; and under the guidance of the high-quality edge information, using an improved joint two-sided filter to carry out interpolation construction on the low-resolution test image so as to obtain an expected high-resolution depth image. By use of the method, the phenomena of edge blur and sawteeth in a reconstruction process can be avoided, the sharpness of the image edge is kept, and in addition, noise can be effectively inhibited.

Description

technical field [0001] The invention belongs to the field of computer vision and image processing, in particular to a method for super-resolution reconstruction of a depth image. Background technique [0002] With the rapid development of computer vision technology, only relying on information in color images can no longer meet the application requirements of computer vision, and depth images can just make up for this defect. Depth images describe distance-related information about objects in a scene. Some active sensors, such as Kinect and PMD (Photonic Mixer Device) cameras, can simply obtain the depth information of the scene, but due to the limitation of the sensor hardware system and the influence of the external environment, the resolution of the directly obtained depth image is low. It affects its requirements in practical applications. Therefore, it is of great significance to improve the resolution of the depth image. [0003] Depth image super-resolution reconst...

AI Technical Summary

Problems solved by technology

[0004] The interpolation method is to infer the pixel to be interpolated by the pixel value of the adjacent position. This method is simple and efficient, but it is easy to cause blurred and jagged edges of the image; the method based on the fusion of the same scene image sequence is to use the relevant low-resolution depth image The information in the image is used to make up for the pixel information of the unknown position. This method has strict requirements on camera movement between adjacent frames. It requires that the lens and the object in the scene only have a slight movement in the direction parallel to the plane of the lens, which is difficult to meet in practical applications; based on The high-resolution color image-guided method is to guide the depth image reconstruction through the consistency of the content distribution of the depth image and the color image, which requires the assistance of a high-resolution color image that is highly registered with the depth image; the method based on example image learning It is necessary to learn a feature relationship between a low-resolution image and a high-resolution image from the example image set, and use this feature relationship to guide image reconstruction. This method has a strong dependence on the example image training set and uses different training data sets. The effect of rebuilding may also be different

Images