A salient object detection method based on attention mechanism

Abstract

An attention-based salient object detection method. It relates to the fields of computer vision and digital image processing, and specifically relates to a salient object detection method based on an attention mechanism. A salient object detection method based on the attention mechanism is provided. Firstly, a top-down attention network is designed to purify the convolutional features of each layer, and then a second-order item is introduced to design a residual feature fusion network to better preserve residual features. Difference feature; the implementation takes any static color image as input, and its output is a saliency map with the same size as the input image. White in the saliency map represents the salient object area, and black represents the background area. The invention can obtain a high-resolution saliency map, and at the same time can better detect small salient objects.

Description

technical field [0001] The invention relates to the fields of computer vision and digital image processing, in particular to a method for detecting a salient object based on an attention mechanism. Background technique [0002] Salient object detection is a popular research topic in the field of computer vision. Its purpose is to extract attractive objects or regions in an image and assign them a saliency value. As a preprocessing step, it can be applied to other high-level vision tasks such as weakly supervised semantic segmentation, object recognition, etc. Traditional salient object detection methods are based on manually designed features, such as contrast, background center difference, etc. These hand-designed low-level visual features are difficult to capture semantic information, and thus do not perform well in complex scenes. In recent years, thanks to the rapid development of deep learning, the performance of salient object detection has been greatly improved. How...

AI Technical Summary

Problems solved by technology

However, the existing salient object detection models are fine-tuned based on the image classification model. Due to the differences in the tasks of the two, the features learned by the image classification network cannot accurately locate the object boundary, and directly using it for salient object detection cannot obtain high resolution. The saliency map of the rate, especially at object boundaries

In addition, in order to be able to detect multi-scale salient objects, it is usually necessary to fuse convolutional features of different scales. However, the existing feature fusion methods simply add or merge, which tends to weaken the residual features, thus affecting the detection of small salient objects.

Images