Design And Implementation Of A Vertical Content-Based Picture Search Engine

The traditional image retrieval system is based on text matching method. It tries to match up the search keyword with graphics presentation, and then returns the corresponding query results to the user. However, this retrieval method has great limitations, because the textual description can’t always express features of image. Content-based image retrieval technology makes up for the deficiency of the text image retrieval. It is focused on the image content, and do analysis through the image information (such as color, shape, texture), and then establish the corresponding image index for the query. However, the existing systems typically employ a single feature, such as color or texture features for image retrieval. It’s less efficient.This thesis studies CEDD (Color and Edge Directivity Descriptor) features combined color with texture characteristics, and proposes the improvement program based on binary Haar Wavelet Transformation for CEDD. On the basis of these researches, it designs and implements the whole search engine system, which have features such as lightweight, cross-platform and scalable. The system achieves Web crawler function, and also gives concrete realization of indexing, retrieval and user interaction module. In addition, the thesis integrates color features ColorLayout and texture features Tamura (index modules), in order to compare the effect of CEDD and retrieval features.In order to verify the characteristics of the retrieval efficiency, the thesis selects a part of Corel image set as test set. Through using the corresponding test methods and evaluation standard, it gives the comparison of search results among the three image features. In the same or similar situation, it achieves achieve contrast with existing systems. Experiments show that, CEDD features have higher retrieval efficiency than original features and can get better results.