Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

A salient object detection method based on deep network layering and multi-task training

An object detection and deep network technology, applied in the field of image processing and computer vision, can solve the problem of insufficient details of the edge of the object

Active Publication Date: 2019-06-21
SICHUAN UNIV
View PDF7 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Although there are many existing detection models and algorithms for salient objects, it is still extremely challenging to detect salient objects from complex and unrestricted scenes. How to more accurately locate salient objects and segment the exact boundaries of the object is the key to be solved urgently one of the problems
At present, the emerging saliency detection method based on deep learning (for example, Li et al. proposed "DeepSaliency: Multi-Task Deep NeuralNetwork Model for Salient Object Detection" in 2016), although using multi-task training, in terms of locating salient objects It has great advantages, but it still has a lot of shortcomings in describing the details of the edge of the object

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
  • A salient object detection method based on deep network layering and multi-task training
  • A salient object detection method based on deep network layering and multi-task training
  • A salient object detection method based on deep network layering and multi-task training

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0033] In the first step, one or more tasks associated with the salient object detection task prepare training pictures required for multi-task training. Among them, for the saliency detection task, the training picture includes the original image and its corresponding saliency map, and for other tasks, the training picture includes the original image and its corresponding real result. Other tasks refer to one or more tasks that are intrinsically related to the saliency detection task and can share features, such as semantic segmentation, human gaze point prediction, etc. For semantic segmentation tasks, the training image includes the original image and the class label map to which the region in the image belongs.

[0034] The second step is to design a deep neural network architecture and loss function with hierarchical features. Specific steps include:

[0035] S2-1: Design the network structure, which includes a main forward network and multiple side paths connected to s...

Embodiment 2

[0041] The difference between embodiment 2 and embodiment 1 is that the task associated with the salient object detection task is the human gaze point prediction task, and the multi-task training pictures adopted in the joint training include the original image and the human eye gaze point prediction task picture, and the hierarchical The loss function corresponding to the human gaze point prediction task in the feature deep neural network architecture is the cross-entropy loss function, and the corresponding hierarchical feature deep neural network architecture is as follows: Figure 4 shown.

[0042] Other methods and steps are the same as in Embodiment 1, and will not be repeated here.

[0043] Based on the deep neural network architecture with hierarchical features in Embodiment 1, other tasks are expanded in the manner of this embodiment, and the formed network architecture should fall within the scope of protection of the present invention.

Embodiment 3

[0045] The difference between embodiment 2 and embodiment 1 is that the task associated with the salient object detection task uses multiple tasks, the multiple tasks are human gaze point prediction tasks and semantic segmentation tasks, and the corresponding deep neural network architecture of hierarchical features Such as Figure 5 shown.

[0046] The first step is to prepare the training pictures required for multi-task training. Among them, for the saliency detection task, the training picture includes the original image and its corresponding saliency map; for the semantic segmentation task, the training picture includes the original image and the class label map to which the area in the picture belongs; for the human gaze point prediction task, the training The picture contains the original picture and the prediction picture of human gaze point.

[0047] The second step is to design a deep neural network architecture and loss function with hierarchical features. Specif...

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 a salient object detection method based on deep network layering and multi-task training, and belongs to the technical field of image processing and computer vision. The methodcomprises the following steps: 1, determining one or more tasks associated with a salient object detection task; 2, selecting a multi-task training picture; 3, carrying out joint training on the deepneural network of the hierarchical features to obtain an optimized deep neural network of the hierarchical features, and adopting a deep neural network architecture of the hierarchical features by the deep neural network of the hierarchical features; And 4, inputting an image into the optimized deep neural network of the hierarchical features to obtain a salient object detection result. Accordingto the method, multi-task joint training is utilized, hierarchical characteristics of the deep neural network are fused, more accurate positioning of salient object detection is realized, and more accurate and detailed object edge description is realized.

Description

technical field [0001] The invention relates to the technical fields of image processing and computer vision, in particular to a salient object detection method based on deep network layering and multi-task training. Background technique [0002] Salient object detection intends to automatically detect eye-catching objects in images or scenes, and the detected areas or objects can be input into subsequent processing modules as regions of interest, in target detection and recognition, image compression, image retrieval, content-based image Editing and other fields have a wide range of applications. Although there are many existing detection models and algorithms for salient objects, it is still extremely challenging to detect salient objects from complex and unrestricted scenes. How to more accurately locate salient objects and segment the exact boundaries of the object is the key to be solved urgently one of the problems. At present, the emerging saliency detection method ...

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
IPC IPC(8): G06K9/00G06N3/08G06N3/04
Inventor 傅可人赵启军
Owner SICHUAN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products