Research On Image Retrieval Algorithm Based On Saliency Object Detection

With the in-depth application of digital technology in all walks of life,a large number of image information has been born,and saliency object detection technology has gradually been applied to the field of image retrieval,although a large number of excellent results have been achieved,there are still problems such as inaccurate extraction of significant areas and low retrieval efficiency.Therefore,for the image retrieval technology based on saliency object detection,this paper uses the traditional algorithm and convolutional neural network respectively to extract the salient area of the image,and fuse it with the original image as a query image,so as to improve the accuracy and efficiency of retrieval.The main work of this article is as follows:1.In order to solve the problem that the salient area is not prominent enough and there is redundant information in image retrieval based on saliency object detection,this paper first improves the saliency target detection FT algorithm by using the single-threshold segmentation and multi-threshold iterative segmentation methods to remove the noise information and improve the saliency map effect.Combined with the mask idea,the weight matrix is customized,and the original image is fused to enhance the weight of the significant area.This method can effectively improve the accuracy of image retrieval,which has certain theoretical and application significance.2.In order to obtain a more precise significant area,this paper uses convolutional neural networks to improve the PoolNet model to improve the effect of salient regions and improve efficiency.Firstly,in order to make the saliency object detection more convenient to apply to image retrieval,the lightweight structure Mobilenetv3 is used to replace the original backbone network,and the coordinate attention mechanism module is introduced to enhance the detailed feature information of the significant area,and the loss function is introduced to optimize the model to improve the accuracy,and finally F1-score and MAE are used to evaluate the accuracy of the model,and FPS is used to evaluate the speed of object detection.Experimental results show that the improved model can improve efficiency while ensuring accuracy.3.An image retrieval system based on saliency object detection was established,and the fusion images obtained by the above two methods were used as input images for retrieval experiments.When the search results are available,the accuracy rate is used as an evaluation index to verify the effectiveness of the two methods,and the experimental results show that both methods can effectively improve the accuracy of retrieval.