Design And Implementation Of A Text-Based Image Retrieval Platform

The text is the symbolic system used by mankind to communicate.It is the crystallization of high intelligence in mankind,and plays an irreplaceable role in the wide range of activities such as knowledge and information transmission.With the explosive growth of Internet documents,images,and voice data,an realistic issue emerges.How to search for images that contains specific keywords from the large-scale internet images,solving this problem will be of great use for people's daily life,work and internet environment purification.In addition,The text information contained in the image can be used as an important clue to understand the image content.The text-based platform can be also used as an auxiliary part of the semantic-based image retrieval.So the Implementation and design of text-based image retrieval platform is full of practical significance.The main content of this thesis is described as following:First of all,we collected the internet image dataset,the internet images mainly consist of natural scenes,electronic scanning and digital synthesis images.Due to its diverse source of collecting ways,our dataset exhibits more variability in terms of color,font,and intensity,compared to single natural scenes.In this thesis,our text-based image retrieval platform will not only evaluated on the open-access natural-scene dataset,but also on internet image dataset collected in our thesis to further verify the performance of the platform.Learning from the other work in the field,we first implements a text-based image retrieval platform based on text detection and text recognition technology.In recent years the text detection and text recognition algorithms have made great progress.The platform combining the advantages of these two mature technologies has achieved good results on the public dataset and the dataset collected in our thesis.Considering the scalability to the large-scale database and the flexibility of retrieval of the image retrieval platform,we presents a text-based image retrieval platform with the help of text detection and texture feature extraction.Taking into account the timeconsumption of the text recognition algorithm,when faced with large-scale dataset,the platform presented before will consume a lot of time to prepare.The platform replaces the original text recognition process with texture feature extraction process,This greatly reduces the preprocessing time of the database and improves the flexibility of retrieval,while maintaining the accuracy of the image retrieval and responding speed.Compared with the retrieval platform based on text detection and text recognition algorithm,This platform will have higher practical value.The thesis solves the problems of current text-based image retrieval platform which is based on text detection and text recognition algorithm,such as large-scale dataset preparation,rigid retrieval keyword and so on.Therefore,the text-based image retrieval platform with the help of text detection and texture feature extraction algorithm,compared to the existing text detection and text recognition platform,is more effective.