Papers by Keyword: Image Retrieval

Abstract: Due to the diversity and complexity of design patent images, it is difficult to retrieve well if extracting features from images directly. A design patent image retrieval method based on Gabor filter and LBP is proposed in the paper. Firstly, doing low-pass filtering to the normalized images with Gabor filter to amplify the images’ details, then extracting image’s texture feature with LBP algorithm, calculating images’ similarity according to the distance formula after feature vectors’ internal normalization, finally return several similar images. The experimental results show that this retrieval method get better retrieval accuracy and correct rate.

Abstract: The Context-Aware Saliency (CA) model—is a new model used for saliency detection—has strong limitations: It is very time consuming. This paper improved the shortcoming of this model namely Fast-CA and proposed a novel framework for image retrieval and image representation. The proposed framework derives from Fast-CA and multi-texton histogram. And the mechanisms of visual attention are simulated and used to detect saliency areas of an image. Furthermore, a very simple threshold method is adopted to detect the dominant saliency areas. Color, texture and edge features are further extracted to describe image content within the dominant saliency areas, and then those features are integrated into one entity as image representation, where image representation is so called the dominant saliency areas histogram (DSAH) and used for image retrieval. Experimental results indicate that our algorithm outperform multi-texton histogram (MTH) and edge histogram descriptors (EHD) on Corel dataset with 10000 natural images.

Abstract: This paper firstly studies the texture features. We construct a gray-difference primitive co-occurrence matrix to extract texture features by combining statistical methods with structural ones. The experiment results show that the features of the gray-difference primitive co-occurrence matrix are more delicate than the traditional gray co-occurrence matrix.

Abstract: With the Rapid Development of Internet, Image Retrieval is more and more Important,For Image Retrieval, it Requires Not only a Real-Time Retrieval Speed but also the Accuracy of the Results.The Method Described in this Article: Firstly,We should Quantify the HSV Color Space of the Image Non-Uniformly,and then Extract the Weighted Color Autocorrelogram ,and Finally Build an Indexing Library Based Locality-Sensitive Hashing.The Method Successfully Solves the Problem of Dimension Disaster.With Different Scales Image Datasets，the Experimental Results Show that the Method can Retrieve Similar Images Quickly with the Suitable Parameters Selected.

Abstract: Hashing which maps data into binary codes in Hamming space has attracted more and more attentions for approximate nearest neighbor search due to its high efficiency and reduced storage cost. K-means hashing (KH) is a novel hashing method which firstly quantizes the data by codewords and then uses the indices of codewords to encode the data. However, in KH, only the codewords are updated to minimize the quantization error and affinity error while the indices of codewords remain the same after they are initialized. In this paper, we propose an optimized k-means hashing (OKH) method to encode data by binary codes. In our method, we simultaneously optimize the codewords and the indices of them to minimize the quantization error and the affinity error. Our OKH method can find both the optimal codewords and the optiaml indices, and the resulting binary codes in Hamming space can better preserve the original neighborhood structure of the data. Besides, OKH can further be generalized to a product space. Extensive experiments have verified the superiority of OKH over KH and other state-of-the-art hashing methods.

Abstract: Due to its fast query speed and reduced storage cost, hashing, which tries to learn binary code representation for data with the expectation of preserving the neighborhood structure in the original data space, has been widely used in a large variety of applications like image retrieval. For most existing image retrieval methods with hashing, there are two main steps: describe images with feature vectors, and then use hashing methods to encode the feature vectors. In this paper, we make two research contributions. First, we creatively propose to use simhash which can be intrinsically combined with the popular image representation method, Bag-of-visual-words (BoW) for image retrieval. Second, we novelly incorporate “locality-sensitive” hashing into simhash to take the correlation of the visual words of BoW into consideration to make similar visual words have similar fingerprint. Extensive experiments have verified the superiority of our method over some state-of-the-art methods for image retrieval task.

Abstract: Product quantization (PQ) is an efficient and effective vector quantization approach to fast approximate nearest neighbor (ANN) search especially for high-dimensional data. The basic idea of PQ is to decompose the original data space into the Cartesian product of some low-dimensional subspaces and then every subspace is quantized separately with the same number of codewords. However, the performance of PQ depends largely on the distribution of the original data. If the distributions of every subspace have larger difference, PQ will achieve bad results as shown in our experiments. In this paper, we propose a uniform variance product quantization (UVPQ) scheme to project the data by a uniform variance projection before decompose it, which can minimize the subspace distribution difference of the whole space. UVPQ can guarantee good results however the data rotate. Extensive experiments have verified the superiority of UVPQ over PQ for ANN search.

Abstract: This paper firstly studies the image color features based on wavelet territory. We introduce a color features’ extract method based on HSI low-frequency subband color features after partition. Firstly, according to the image attention from human eyes, we split the image into sub-blocks. Then extract HSI low-frequency subband color features of each sub-block after wavelet transform, and we can obtain the image color features by weighting. Comparing with traditional histogram method, the experiment results show that the proposed algorithm based on weighted dominant color feature has better retrieval precision.

Abstract: This paper introduces a HVS Weighted color features’ extract method. Firstly, we split the image into sub-blocks and draw the color feature consists of dominant colors in each sub-block. Then weighting the gained color features by making use of Human Visual System. So we can obtain the weighted dominant color feature. Comparing with traditional histogram method and split blocks dominant color method, the experiment results show that the proposed algorithm based on weighted dominant color feature has better retrieval precision.

Abstract: Content-based image retrieval (CBIR) system can be used to effectively and precisely retrieve the desired images from a large image database, and the development has become an important research issue.Statistical methods like, gray level co-occurrence matrix (GLCM) and the autocorrelation function are used to extract texture feature. Region-based methods utilize information from both boundaries and interior regions of the shape. Shape features like perimeter, area, centroid, circularity, solidity based on region can be extracted in the feature space. Similar images can be retrieved using minimum distance classifiers with and without clustering algorithm .Time complexity and the retrieval efficiency has been analyed and compared on both the methods. The experiments have been conducted on MPEG-7 dataset.