Indoor RGB-D image semantic segmentation method based on wavelet transform

Abstract

The invention discloses an indoor RGB-D image semantic segmentation method based on wavelet transform. The method comprises the following steps: firstly, constructing a convolutional neural network based on wavelet transform; secondly, performing multi-scale fusion on image features of different frequencies of the original image at the joint of the encoder and the decoder through a wavelet transform and inverse transform combined module; and finally, carrying out multiple times of up-sampling on the features through a decoder network, carrying out double up-sampling on the features by each module of a decoder, better carrying out feature mapping through convolution and jump connection of an encoder, gradually recovering a high-resolution image, and outputting a semantic segmentation result. According to the method, high-frequency features which are not utilized by a conventional method are fully utilized, information such as contour details in image information is reserved, the indoor image semantic segmentation accuracy and the reasoning speed are effectively improved, and the parameter quantity of a network is reduced.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to an indoor RGB-D image semantic segmentation method based on wavelet transform. Background technique [0002] With the development of the computer vision field, semantic segmentation of images has become an important topic in this field, classifying each pixel in the image and predicting the shape of its label and position, and providing the relevant scene through the above operations. complete understanding. Semantic segmentation is now widely used in autonomous driving, remote sensing analysis, and medical image processing. For indoor scene semantic segmentation, due to the complex factors affecting indoor scene semantic segmentation (lighting, occlusion, etc.), reducing the influence of these factors in indoor scene research and improving the accuracy of semantic segmentation has become the main problem. . Previous deep learning methods achieve end-to-end image semantic segmen...

AI Technical Summary

Problems solved by technology

However, there are still many problems in the above method: 1) The depth image is simply fused with the color image as the fourth channel, and the complementarity between RGB color information and depth information is not fully utilized; 2) There are problems such as loss of multiple scale information features in different methods, However, for the indoor scene semantic segmentation problem, the efficiency of multi-scale feature extraction affects the segmentation accuracy of small target objects

3) The depth image is more sparse than the RGB image, which represents the depth value of each pixel in the RGB image, and the edge contour information, that is, high-frequency information, can be better obtained from the depth image, but traditional convolution and pooling Such operations tend to lose high-frequency features

Images