Unlock instant, AI-driven research and patent intelligence for your innovation.

RGB-D salient target detection method using adaptive feature fusion

A technology of RGB-D and feature fusion, applied in the field of image processing, can solve the problem of low segmentation accuracy, inability to effectively integrate RGB and Depth, etc., to achieve the effect of improving the effect

Pending Publication Date: 2021-10-22
HANGZHOU DIANZI UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The existing RGB-D salient target detection method cannot effectively integrate the features of RGB and Depth data streams in the decoder stage, resulting in low segmentation accuracy

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • RGB-D salient target detection method using adaptive feature fusion
  • RGB-D salient target detection method using adaptive feature fusion
  • RGB-D salient target detection method using adaptive feature fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0028] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0029] A RGB-D salient object detection method using adaptive feature fusion proposed by the present invention uses a computer to automatically segment out the salient objects in RGB-D for an image.

[0030] Such as figure 1 Shown, method of the present invention comprises the following steps:

[0031] The specific method of step (1) is as follows:

[0032] For RGB and depth figure 2 data streams, each using two Resnet34 networks pre-trained on ImageNet to encode its features. The network parameters of Resnet34 are shown in Table 1. For the two data streams, the output features of conv2_x, conv3_x, conv4_x, and conv5_x are taken out as the output of the four levels of the encoder. It is known that the input feature of the RGB image is (H, W, 3) (the first two numbers in the brackets indicate the resolution, and the last number indicates the number of ch...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses an RGB-D salient target detection method using adaptive feature fusion, which comprises the following steps: firstly, encoding two data streams, namely an RGB image and a Depth image, by using two encoder networks with the same structure, and obtaining m-level features for the two data streams after encoding; then, respectively using adaptive feature fusion in the decoding process of the RGB data stream and the Depth data stream; fusing the features of the two data streams in the joint data stream; and finally, optimizing the network model by using a loss function. According to the method provided by the invention, the RGB-D salient target detection effect can be improved. By using adaptive feature fusion, features of different levels in a data stream can be effectively fused. The features of two data streams can be effectively fused by using an inter-data stream feature fusion method.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an RGB-D salient object detection method. Background technique [0002] Image salient object detection refers to the automatic segmentation of salient objects in images by computer. The technology has a wide range of application scenarios, such as image compression, or as a preprocessing task for object recognition, semantic segmentation, object tracking, etc. [0003] In order to obtain accurate image salient object segmentation results, the traditional solution is to manually design a feature extraction method, and then perform pixel-by-pixel classification based on this feature. [0004] The existing RGB-D salient target detection method cannot effectively integrate the features of RGB and Depth data streams in the decoder stage, resulting in low segmentation accuracy. Therefore, this paper proposes a method that can effectively fuse the features of the two data str...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/10G06K9/46G06K9/62
CPCG06T7/10G06T2207/10016G06T2207/10024G06F18/253Y02T10/40
Inventor 张继勇吕成涛颜成钢孙垚棋李宗鹏
Owner HANGZHOU DIANZI UNIV