Research On Intelligent Optimization Algorithm And Its Application In Image Retrieval

Intelligent optimization algorithm is inspired by animal population behavior, human social behavior or natural phenomena. Intelligent optimization algorithm is simple, robust, suitable for parallel computing and easy to achieve global optimization when compared with traditional optimization algorithm. As a result, scholars have paid great attention to the algorithm. Brain Strom Optimization (BSO) is the first algorithm proposed by inspiring human’s creative thinking. It has the advantages of simplicity and high searching precision. In recent years, BSO has been widely applied to the communication, control and military fields. However, BSO is newly developed and it is easy to trap into local optimal.To overcome the shortcomings of BSO, Niche Brain Strom Optimization algorithm (NBSO) and Artificial Immune Brain Strom Optimization algorithm (AIBSO) is proposed in the paper. BSO uses clustering operator and it results in the low diversity of population. To maintain the population diversity, niche is introduced in the NBSO. Individuals update in different living environments rather cluster in a living environment, so that individuals distribute separately in the available area. NBSO maintains the diversity of population effectively and prevents premature convergence. AIBSO has both the properties of the origin brain storm optimization algorithm and the immune mechanism of immune system, can maintain the population diversity and improve the global searching ability. Meanwhile vaccination strategy is introduced to improve the speed and precision of evolution. Experiment results show the effectiveness of these two improved BSO algorithms.With the fast development of information technology and increasing number of image database, how to retrieval a large amount of information from the image quick and effective becomes more and more important. BSO is simple, robust and has high searching precision. AIBSO is applied into the image retrieval. Content-based image retrieval (CBIR) extracts the color, texture, shape and other low-level visual features of images to realize image matching. Compared with image retrieval based on single feature, image retrieval based on multi-feature fusion can fully represent image information. In the multi-feature fusion, the ratio of each feature selection is critical to the search result. Traditional method is manually set or proportional integration, which ignores the priority between the various features. This paper uses AIBSO for image retrieval. Color histogram, color moment, color structure descriptor, Tamura texture feature, GLCM texture, texture wavelet transform, Gabor transform texture, edge histogram descriptors and Hu invariant moment are extracted in the paper and AIBSO is used for image retrieval. Experiment results show that the proposed method can retrieval the target image accurately and improve the precision of the system.In CBIR, feature selection has a significant impact on the performance of image retrieval. However, manually selection is a very laborious method and requires expertise. Depth study find Distributed characteristic of data through forming a more abstract level characterized by a combination of low-level features, it extracts features directly from the original image and there is no need to extract features manually. In the paper, the depth of convolution neural network is applied to extract image features automatically, and the extracted features are used for image retrieval, which reduces the influence from different features.