Article Preview
TopIntroduction
With the rapid development of science and technology, artificial intelligence industry has experienced rapid development for more than ten years. At present, China's AI technology is developing rapidly, and the country should set reasonable short-term goals for AI promotion policies, and start to pragmatically develop AI infrastructure from three aspects: improving the system, making up for the shortcomings and improving the regulation(Zhang Xin & Wang Ming hui, 2019). AI is applied to various industries, such as social and smart homes(Kar, 2022; Sharma et al., 2022). Machine learning, as a part of the theme of artificial intelligence, has brought significant changes to the industry and made a significant contribution to the detection of phishing websites(Almomani et al., 2022; Madhu et al., 2022). The image retrieval technology used in this paper is a part of machine learning. Existing image retrieval mainly focuses on the application of convolution neural networks in image representation and classification(Wei Mingzhu et al., 2021). Image retrieval has gone through two phases, manual and depth-based, with the goal of finding and querying images from database images that contain the same instance(Zhang Hao & Wu Jianxin, 2018). With the development of image retrieval technology, people require more and more efficient retrieval. Many image retrieval systems based on CNN have an insufficient expression of image features, lack accuracy and robustness(Wei Yun & Yan Zhengyi, 2021). Image retrieval based on in-depth learning extracts the overall image features by convolution neural network and retrieves the similarity of the overall image features, converts image features into corresponding feature vectors. which results in low accuracy and poor robustness of the retrieval results of human image background similarity.
Image retrieval technology mainly consists of several steps: input pictures, feature extraction, measurement learning, and reordering. Feature extraction is a crucial step to determine the accuracy and robustness of image retrieval, the content is to reduce the dimension of the image data, extract the discriminant information of the data, and reduce a picture to a vector. However, the feature extraction of the whole picture reduces the feature of the whole picture to a vector, which results in low accuracy and poor robustness in image background similarity retrieval. Background similarity retrieval of portraits based on background analysis can improve the accuracy of background similarity retrieval by dividing portraits into different parts and leaving background information during the feature extraction stage.