Research On Image Retrieval Theory And Several Key Technologies For Digital Learning

With the popularity of educational information,the construction of digital learning resources has been paid more and more attention,and it is regarded as the cornerstone of the development of educational technology.Digital images are one of the most important components of digital learning resources.In order to solve the complicated and large-scale digital image resources,how to find the target image accurately and quickly is one of the problems to be solved urgently in the construction of digital learning resources.The content-based image retrieval technology(CBIR)provides a feasible solution to the problem.In this paper,the theory and method of image retrieval based on content are studied by using statistical modeling,robust interest point extraction and related feedback learning,and some research results are obtained:1.A texture image retrieval algorithm based on multi-statistical model combinatorial modeling is proposed.The algorithm is used to decompose the image by PDTDFB.The amplitude and relative phase of the coefficients of the high frequency sub-band are decomposed.Then Weibull and Vonn distribution probability density functions are used to statistically model them separately.Two sets of parameters will be used as a texture feature;the final calculation by similarity,the results will be returned to the user.The simulation results show that the proposed method can reduce the feature dimension effectively by using the parameter group as the image feature.The hybrid modeling avoids the missing of the texture information and obtains the ideal retrieval effect on the basis of low time complexity.2.A local feature color image retrieval algorithm based on robust points of interest is proposed.The algorithm first constructs the color invariant image.And then use the improved high-order derivative SIFER(Scale-Invariant Feature Detector with Error Resilience)to extract the robust feature points of the image.Then we calculate the local angle phase(LAP)of the feature points and construct the LAP histogram as the texture of the image feature.The K-means algorithm is used to cluster the obtained feature points,and the color space histogram is constructed as the color feature of the image.Through the normalization process,parameter selection experiment and similarity calculation,the final return to the user search results.Simulation results show that the improved SIFER operator is more sensitive to the region of interest of the image,and can extract the feature points accurately and has strong robustness.The use of local feature points achieves a reduced dimension effect,with low timecomplexity and an ideal retrieval effect.3.An improved feedback image retrieval algorithm based on improved support vector machine is presented.The algorithm first uses a kind of efficient active learning algorithm in the feedback annotation to achieve the problem of reducing the workload while alleviating the positive and negative cases of the feedback results.The K-Medoids algorithm is used to reduce the clustering time and improve the classification effect.The use of the Weighted Extreme Support Vector Machine(WESVM)classifier allows the error terms of different samples to be best suited to their own penalty factors,thus eliminating interference from irrelevant noise and outliers.Finally,the parameters of the classifier are optimized by the combination of intelligent algorithm and grid algorithm.The simulation results show that the human-computer interaction algorithm which combines the optimization feedback process and the sample labeling process effectively solves the problem of "semantic gap" and obtains better retrieval results in fewer feedback times.