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

Semantic segmentation method based on background blurring

A technology of semantic segmentation and background blurring, applied in the field of semantic segmentation based on background blurring, to achieve the effect of improving network segmentation performance

Pending Publication Date: 2022-01-28
河南奇点网络科技有限公司
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, from semantic segmentation to real-time semantic segmentation, from redundancy to streamlined network architecture, existing improvements generally achieve better segmentation by designing and improving the structure of the network itself, and using a large number of data enhancement methods, while ignoring the data set itself. The impact of features on segmentation results

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
  • Semantic segmentation method based on background blurring
  • Semantic segmentation method based on background blurring
  • Semantic segmentation method based on background blurring

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach 1

[0024] Embodiment 1: This semantic segmentation method based on background blur is implemented by embedding a BOKEH module in the existing FCN segmentation network, and the algorithm of the BOKEH module is described as Among them, img(:,:,k) represents the pixel value size on channel k of the picture; σ represents the blur factor; R B* Indicates the background rate of the overall training set of the data set; R B Represents the proportion of the background domain; BGL represents the background position information of the image.

[0025] said in, Indicates domain of interest; Indicates the bokeh domain. said Among them, 0≤i≤H-1, 0≤j≤W-1, k=0,1,2, σ∈(0,1]; img(i,j,k) means that the picture k channel (i, j) The size of the pixel value at the position; * indicates the matrix Hadamard product; img(i,j,k)*BGL(i,j) indicates that the k channel domains of img(i,j,k) are respectively related to BGL(i,j ) to do Hadamard product. The σ selection strategy is σ=-R B × R B...

Embodiment approach 2

[0038] Implementation mode 2: if Figure 5 As shown, specific to the FCN network model, the FCN network described in the semantic segmentation method based on background blurring includes an input terminal, a convolution unit, a pooling unit, an encoder unit, a decoder unit, and an output terminal, wherein the BOKEH The module is embedded between the input and the first convolutional unit. The BOKEH module is as described in Embodiment 1 and will not be described again.

[0039] Comparative analysis of technical effects:

[0040]1. Using the classic FCN network and the real-time bilateral network BiSeNet.

[0041] In order to more accurately capture the impact of background information on segmentation accuracy, all data augmentation methods except crop size are removed. Denote as (Re)FCN-8s, (Re)FCN-16s, (Re)FCN-32s and (Re)BiSeNet, respectively.

[0042] 2. Apply the method to multiple segmentation networks on two datasets and report the results on both datasets. specifi...

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 relates to a semantic segmentation method based on background blurring, which is realized by embedding a BOKEH module in an existing FCN semantic segmentation network, and the algorithm description of the BOKEH module is as follows: img (:,:, k) represents the pixel value on a picture k channel; sigma represents a blurring factor; RB * represents the overall background rate of the data set training set; RB represents a background domain ratio; the BGL represents background position information of the image. According to the semantic segmentation method based on background blurring, a data set is deeply analyzed and optimized, and on the premise that the framework of an existing network is not changed or data is not increased, the network segmentation performance is effectively improved by embedding a background blurring module BOKEH in an existing FCN semantic segmentation network.

Description

technical field [0001] The invention relates to the technical field of image segmentation, in particular to a semantic segmentation method based on background blurring. Background technique [0002] In recent years, deep convolutional neural networks have been increasingly used to implement image pixel-level, end-to-end segmentation tasks—semantic segmentation. As an important part of computer perspective tasks, semantic segmentation is widely used in the fields of autonomous driving, robot perception, reality augmentation, and video surveillance. [0003] The emergence and application of FCN (Fully Convolutional Neural Network) has greatly simplified the traditional methods used to solve semantic segmentation problems. Based on the existing data sets, the segmentation results are relatively high, or even saturated. However, from semantic segmentation to real-time semantic segmentation, from redundancy to streamlined network architecture, existing improvements generally ach...

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): G06V10/26G06V10/774G06V10/82G06V20/70G06K9/62G06N3/04G06N3/08
CPCG06N3/082G06N3/045G06F18/214
Inventor 谌东宇刘长江李豪苏跃斌
Owner 河南奇点网络科技有限公司