Image emotion classification method based on LSTM network and attention mechanism

A technology of emotion classification and attention, applied in the field of image processing, can solve the problem of low precision and achieve the effect of reducing the impact of semantic gap

CN110263808AActive Publication Date: 2019-09-20GUANGDONG UNIV OF TECH
4 Cites 4 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Publication Date
2019-09-20

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses an image emotion classification method based on an LSTM network and an attention mechanism. The method comprises the steps of initializing an original image; setting an image emotion label classifier based on an LSTM network and an attention mechanism; training an image emotion label classifier; testing an image emotion label classifier; and performing sentiment classification on the target image by using the tested image sentiment label classifier to obtain a sentiment classification result. According to the method, image emotion attributes are introduced, an attention mechanism acts on a visual feature map of an image, weighted features of the emotion attributes and the visual features are obtained through calculation and serve as the initial state of an LSTM network, and therefore image emotion areas are accurately detected, and meanwhile multiple emotion areas of the image are focused as much as possible; and based on the image emotion regions, emotion classification is carried out on the image through a classifier, so that image emotion prediction is more accurate, and an image emotion classification result obtained through the method conforms to a human emotion standard.
Need to check novelty before this filing date? Find Prior Art

Description

Technical field

[0001] The present invention relates to the technical field of image processing, and more specifically, to an image emotion classification method based on an LSTM network and an attention mechanism. Background technique

[0002] At present, people at home and abroad have begun to research and explore image emotion classification. At present, the general way of image emotion classification is to select the image to be studied, extract the visual features of the image, establish the emotional space, and select the appropriate classifier to study. The images are trained first and then classified. However, in the visual task of image sentiment analysis, the attention system that affects humans is often the local area of ​​the image rather than the overall area of ​​the image, and the existing image sentiment classification model is mainly based on the overall area of ​​the image, which leads to the unsatisfactory effect of emotion classification. . Summary of the in...

Examples

Embodiment 1

[0054] An image sentiment classification method based on LSTM network and attention mechanism, such as figure 1 , 2 As shown, including the following steps:

[0055] S1. Original image initialization: Obtain the original image from the image emotion database, divide the original image into a training image, a test image, and a target image, and initialize the original image to generate a corresponding image target area; Each of the original images corresponds to an emotional attribute and an emotional label; each image in the data set corresponds to an emotional attribute and an emotional label. This embodiment 1 uses the vso image emotion database, in which each picture corresponds to an emotion attribute and an emotion label; image 3 As shown, the happy baby in the upper left of the figure has an emotional attribute of happy and an emotional label of positive.