Oracle Image Retrieval Method Based On Siamese Neural Network

The Oracle Bone Inscription is the oldest hieroglyph in China and plays an important role in the long history of the development of Chinese characters.However,it takes a lot of time and effort to identify uninterpreted words or collect and classify relevant example content.The use of image retrieval technology can accurately find a similar image in a large amount of image big data with a small space-time overhead.Image retrieval has also become an important research hotspot in the field of multimedia and information retrieval in recent years.Therefore,how to use image retrieval technology to classify and recognize Oracle images is also of great significance.The essence of image retrieval is the process of characterizing the image and then calculating the similarity.However,the traditional structural features or image processing methods for extracting features are affected by Oracle’s time,region,and engraving strength.The same single-word Oracle has a variety of glyphs.Poor generality of the method,poor feature identification ability,etc.,resulting in low retrieval accuracy;while traditional deep learning methods rely heavily on the data volume of training data sets,from the perspective of computer vision,there are few studies on oracle bone character analysis,and the data distribution is scattered,And there is no related complete large public data set.Therefore,this paper adopts a Siamese Neural Network model that is more suitable for multi-classification and small sample data,which can overcome the above problems of poor generality and few data sets.And VLAD can obtain effective statistical information about the local descriptors aggregated on the image,improving the effectiveness of features.Therefore,the model is composed of two parallel branches,and each branch is composed of CNN and VLAD serially.The image extracts the depth descriptor through the CNN layer,and then uses VLAD to generate the final representation of the Oracle image.The main work contributions of this article can be divided into the following aspects:(1)This article uses the crawler tool to collect nearly 40,000 text pictures and tags of Oracle from the website,which is also one of the larger Oracle data sets as far as we know.(2)This paper proposes an Oracle image retrieval method based on the Siamese Neural Network,which uses the CNN+VLAD method to generate the representation features of the Oracle image.Compared with other methods,the effectiveness of this method is proved through experiments.(3)This article builds an image retrieval system based on the web platform,so that the retrieval method proposed in this article can be applied in practical work.