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

Method and apparatus for detecting salient objects in image

An object detection and image technology, which is applied in image enhancement, image analysis, image data processing, etc., can solve the problems of messy feature maps, large differences in saliency detection images, and lack of regularity in feature maps, so as to reduce complexity and improve effect, the effect of reducing the amount of computation

Active Publication Date: 2019-01-01
NANKAI UNIV +1
View PDF8 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, since the saliency detection images of the side output feature maps of different side output layers are quite different (the feature maps of the shallow layer are too messy, and the feature maps extracted by the deep layer lack regularity), therefore, the side output feature maps of different levels are directly integrated in the fusion layer. If the output feature map is simply fused, then the saliency detection result of the final image to be processed is not ideal

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
  • Method and apparatus for detecting salient objects in image
  • Method and apparatus for detecting salient objects in image
  • Method and apparatus for detecting salient objects in image

Examples

Experimental program
Comparison scheme
Effect test

example 1

[0099] Example 1: Portrait Segmentation

[0100] 410. Using the portrait segmentation data set to train a convolutional neural network model.

[0101] The convolutional neural network model can be as image 3 shown.

[0102] The portrait segmentation data set includes portrait pictures (pictures containing portraits) and the real saliency distribution maps corresponding to the portrait pictures. In addition, in order to increase the training effect, the image can be mirrored, rotated, and the lighting can be changed to avoid over-fitting when training the convolutional neural network.

[0103] 420. For the input portrait picture I h , first down-sampling to obtain a low-resolution image I l , and then process the down-sampled low-resolution image through the trained convolutional neural network, and finally output the low-resolution portrait segmentation image S l .

[0104] The input portrait image is first down-sampled, which can reduce the resolution of the image and ...

example 2

[0111] Example 2: Vehicle Segmentation

[0112] 510. Using the vehicle segmentation data set to train the convolutional neural network model;

[0113] 520. For the input road scene picture I h , first down-sampling to obtain a low-resolution image I l , and then process the down-sampled low-resolution image through the trained convolutional neural network, and finally output the low-resolution vehicle segmentation image S l .

[0114] 530. Segment image S for vehicles l Perform upsampling to obtain a picture S with the same size as the original road scene picture h ;

[0115] 540. According to the road scene picture I h For picture S h Guided filtering is performed to obtain the final vehicle segmentation image.

[0116] It should be understood that the above are only two scenarios for the specific application of the image salient object detection method of the embodiment of the present application. In essence, the image salient object detection method of the embodimen...

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

A method and an apparatus for detecting salient objects in an image are provided. The method comprises the following steps of: respectively performing convolution processing corresponding to at leasttwo convolution layers on an image to be processed to obtain at least two first characteristic maps of the image to be processed, wherein the resolutions of the at least two first characteristic mapsare smaller than the resolutions of the image to be processed, and the resolutions of any two first characteristic maps in the at least two first characteristic maps are different; performing superimposition processing on at least two first feature maps contained in a superimposed set of at least two sets, so as to obtain at least two second feature maps of an image to be processed, wherein at least two sets correspond to different resolutions respectively, at least two sets correspond to at least two second feature maps one by one, and the resolution of the first feature map contained in thesuperposition set is less than or equal to the resolution of the second feature map corresponding to the superposition set; At least two second feature map are stitched together to obtain a saliency map. The present application can improve the effect of significant object detection.

Description

technical field [0001] The present application relates to the field of computer image processing, and more specifically, to a method and device for detecting salient objects in an image. Background technique [0002] Salient object detection is the process of detecting the object region that can most attract the visual attention of the human eye from the image. The existing image salient object detection method is based on the existing convolutional neural network architecture, and fine-tunes the convolutional neural network architecture to achieve salient object detection in images. Specifically, such as figure 1 As shown, each convolutional layer in the convolutional neural network architecture is connected to a side output layer, and all side output layers are connected to a fusion layer. When processing images, the image to be processed outputs feature maps of different resolutions after passing through the convolutional layer. Next, the feature maps of different resol...

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): G06T5/10G06T5/50G06N3/08G06K9/46
CPCG06N3/08G06T5/10G06T5/50G06T7/12G06T2207/20081G06V10/462G06V2201/07G06T7/00G06V10/454G06V10/40G06V30/194G06V30/19173G06F18/214G06F18/2136
Inventor 侯淇彬程明明白蔚周迅溢
Owner NANKAI UNIV