Research And System Implementation Of Multi-Feature Based Re-ranking Image Retrieval

In recent years,content-based image retrieval has attracted great attention in the community of image engineering.Currently,content-based image retrieval has been widely used in electronic commerce,military reconnaissance,video surveillance,and medicalimage processing.To improve the accuracy of image retrieval,we focus the study of the content-basedimage retrieval theory,Then a new multi-feature based re-ranking image retrieval prototype system was implemented for the evaluation of the proposed algorithm.The main contents of this thesis are summarized as follows:1.A new color-based retrieval algorithm is proposed by introducing Krawtchouk chromaticity distribution moments.The image is transformed from RGB space to opponent chromaticity space for the color description by the statistics of image moment vectors.According to the experimental results,the proposed algorithm can reduce feature dimension effectively,promote retrieval accuracy and often outperform traditional HSV-based methods in most cases.2.An image retrieval method based on the self-feedback of multi-features is implemented.By introducing Krawtchouk chromaticity distribution moments,the united method of Krawtchouk moments and Gabor wavelet adopts the self-feedback technique to conquer the drawback of low efficiency led by the multi-features based image retrieval methods.Experiments show that this method effectively improves the image retrieval efficiency.3.An image retrieval re-ranking prototype system based on multi-features is designed in the experiments.The system first selects several candidate image set by SURF-BoVW method,then employs multi-features for image retrieval re-ranking.Experimental results show that the proposed algorithm can achieve higher precision of image retrieval and has better efficiency in most cases.