Depth image recovery method based on improved bilateral filters

Abstract

The invention discloses a depth image recovery method based on improved bilateral filters. The method includes the steps that firstly, the local self-adaption bilateral filter is used for carrying out smooth denoising on an original depth image; then error depth pixels in the depth image are corrected; finally, an RGB image corresponding to the depth image is combined, holes of the depth image are filled through the selective combined bilateral filter, the noise of the finally recovered depth image is effectively restrained, the error depth image pixels are effectively corrected, and a clear depth boundary is provided. The local nonuniform noise in the depth image is effectively restrained, the holes in the depth image are filled, the edges of the recovered depth image are clear and orderly, and later three-dimensional scene recognition, reconstruction and the like are facilitated.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a RGB-D image, that is, a depth image restoration method of a RGB image corresponding thereto. Background technique [0002] With the birth of the new depth measurement technology, it has become very easy to obtain the depth image of the scene, and the measurement device is also very cheap, and the RGB image of the scene corresponding to the depth image can be obtained at the same time, which together with the depth image is called RGB-D image. However, due to the simple measurement principle of this type of device, there are also problems in the quality of the obtained depth image, mainly including: local uneven noise, void phenomenon and wrong depth pixels. At present, the depth image restoration methods obtained by such devices mainly include two categories: (1) only use the depth image information for restoration, (2) combine the depth image and the correspo...

AI Technical Summary

Problems solved by technology

Although the above two methods have high computing speed, they only use depth information, so the effect on depth image restoration is very limited. They are only suitable for algorithms that do not require high quality depth images, and can be used as simple preprocessing algorithms.

Y.Jingyu et al. proposed in 2012 ("Depth recovery using an adaptive color-guided auto-regressive model") to use the energy minimization method to first interpolate and fill the holes in the image, and then iteratively optimize the energy function to achieve hole filling and the effect of denoising, this method uses the RGB-D image information of the current frame, and its defect is that the edges of the restored depth image are very blurred; Massimo Camplani et al. proposed in 2012 (“Accurate depth-color scene modeling for 3d contents generation with low cost depth cameras") use multi-frame RGB-D images to build a scene model, and use the model for depth image restoration, but this method requires the scene in the image to be static, which greatly limits its applicability

[0004] However, the above methods also have a common defect that they do not emphasize the problem of wrong depth pixels and propose a good solution

Images