Image retrieval using color histograms generated by Gauss mixture vector quantization

Gauss mixtures are a popular class of models in statistics and statistical signal processing because they can provide good fits to smooth densities, because they have a rich theory, and because they can be well estimated by existing algorithms such as the EM (expectation maximization) algorithm. We here extend an information theoretic extremal property for source coding from Gaussian sources to Gauss mixtures using high rate quantization theory and extend a method originally used for LPC (linear predictive coding) speech vector quantization to provide a Lloyd clustering approach to the design of Gauss mixture models. The theory provides formulas relating minimum discrimination information (MDI) for model selection and the mean squared error resulting when the MDI criterion is used in an optimized robust classified vector quantizer. It also provides motivation for the use of Gauss mixture models for robust compression systems for general random vectors.

With the development of the internet, more and more images appear in the internet. How to effectively retrieve the desired image is still an important problem. In the past, traditional color histogram is used image retrieval system, but color histograms lack spatial information and are sensitive to intensity variation, color distortion and cropping. As a result, images with similar histograms may have totally different semantics. So the spatial information should be included in color histogram. The color histogram based on saliency map approach is introduced to overcome the above limitations. In this paper, we present a robust image retrieval based on color histogram of saliency map. Firstly, in order to extract useful spatial information of each pixel, the steady saliency map of the images is extracted. Then, color histogram based on saliency map is introduced, and the similarity between color images is computed by using the color histogram of saliency map. Experimental results show that the proposed color image retrieval is more accurate and efficient in retrieving the user-interested images.

We examine the performance of several neural network vector quantization (VQ) methods on image coding. The VQ methods we look at are Kohonen's Self-Organizing Feature Map (KSOFM), Frequency-Sensitive Competitive Learning (FSCL), and Self-Creating and Organizing Neural Network (SCONN). We also look at variations of these algorithms by combining different methods. Our simulation results show that the best performance is achieved by the SCONN and the combination KSOFM and FSCL.

Color histograms are widely used for content-based image retrieval because they are trivial and effective to compute, not sensitive to size and orientation, and robustly tolerate movement of objects in the image and changes in camera viewpoint. However, a color histogram only records an image's overall color compositions and no any spatial information is included, and also retrieval precision is limited. In this paper, a color-spatial histogram, which incorporates spatial information with color compositions without sacrificing the robustness of color histograms, is proposed. The purpose is considering the representation role of color compositions and spatial information in the meantime. Each entry in a color-spatial histogram is the color frequency and corresponding position information of the virtual boundary. Experimental evidence suggests that this new color-spatial histogram outperform not only the traditional color histogram method but also the other color-spatial histogram methods for image retrieval.

Color histogram is widely used for image indexing in content-based image retrieval (CBIR). A color histogram describes the global color distribution of an image. It is very easy to compute and is insensitive to small changes in viewing positions. However, the histogram is not robust to large appearance changes. Moreover, the histogram might give similar results for different kinds of images if the distributions of colors are same in the images. On the other hand, color correlogram is efficiently used for image indexing in content-based image retrieval. Color correlogram extracts not only the color distribution of pixels in images like color histogram, but also extracts the spatial information of pixels in the images. The characteristic of the color Correlogram to take into account the spatial information as well as the distribution of color pixels greatly attracts the researcher for content based image retrieval. In this paper, we propose the image bin (histogram value divisions) separation technique followed by extracting maxima of frequencies and plotting a correlogram. At first, the histogram is first calculated for an image. After that, it is subdivided into four equal bins. Each bin is subdivided into four more bins and for every such subdivision the maxima of frequencies s calculated. This information is stored in the form of a correlogram. The distance between correlogram of the query image with the corresponding correlogram of database images is calculated. The proposed algorithm is tested on a database comprising a large number of images.

As the use of digital media increases, effective retrieval and management techniques become more important. In order to extract useful information from this huge amount of data, many content based image retrieval (CBIR) systems have been developed. A typical CBIR system captures image features such as color, texture, or shape of objects in the query image and tries to retrieve images from the database with similar features. The main advantage of CBIR systems with relevance feedback is that these systems take into account the gap between the high-level concepts and low-level features and subjectivity of human perception of visual content. In this paper, a CBIR system based on color histogram has been presented and a new approach for image storage and retrieval called association-based image retrieval (ABIR) has been studied. The results of both systems have been shown.

Keyblock, which is a new framework we proposed for content-based image retrieval, is a generalization of the text-based information retrieval technology in the image domain. In this framework, keyblocks, which are analogous to keywords in text document retrieval, can be constructed by exploiting the Vector Quantization (VQ) method which has been used for image compression. Then an image can be represented as a code matrix in which the elements are the indices of keyblocks in a codebook. Based on this image representation, information retrieval and database analysis techniques developed in the text domain can be generalized to image retrieval. In this paper, we propose new models named N-block models which are the generalization of the N-gram models in language modeling to extract comprehensive image features. The effort to capture context in a text document motivated the N-gram models. Similarly, the attempt to capture the content in an image motivates us to consider the correlations of keyblocks within an image. By comparing the performance of our approach with conventional techniques using color feature and wavelet texture feature, the experimental results demonstrate the effectiveness of these N-block models.

In this study, different vector quantization methods are used to compress EMG signals. EMG is a versatile biosignal used in many fields of medical research and its compression is necessary for long-term recordings and narrow-band wireless transmission. The quality of lossy compression-decompression process of a biosignal has to be evaluated with the fidelity of essential medical parameters in mind. In this study, the analysis is focused on the preservation of primarily used spectral variables, namely mean frequency and median frequency. Modified versions of the basic vector quantization, classified vector quantization with two codebooks, classified vector quantization with three codebooks, classified vector quantization with four codebooks, gain-shape vector quantization with unquantized gain value, gain-shape vector quantization with eight gain values, and gain-shape vector quantization with four gain values have been implemented and tested with the segment length of 8 samples and the codebook sizes of 32 codevectors and 64 codevectors. The gain values, used both for criteria of codebook selection in classified vector quantization, and as product code in gain-shape vector quantization, was calculated as a sum of absolute segment values. The gain-shape method with eight quantized gain values shows best results in terms of the compression ratio. The general results reveal systematic bias up to 17.34 Hz in the mean frequency value, but a linear model for correcting the error is presented

The monochromatic static image can be represented as a scalar‐valued function f(u, v), for the brightness or density at point (u, v) on the two‐dimensional coordinate space T. For the color image, the image is represented by a vector‐valued function composed of such signal components as red, green, blue or luminance, chrominance and saturation. The sampling of an image can be considered as derivation of a sequence {fi} of spatially discrete sampled values by dividing the coordinate space T. Denoting the partitioning of the coordinate space by {Si}, the sampled value fi is a scalar or vector value, which is determined uniquely by the subimage {f(u, v) |(u, v) ϵ Si}.This paper proposes a new variable rate sampling method which partitions the multidimensional coordinate space efficiently based on the vector quantization method. A high‐speed region partitioning algorithm is also shown which is based on the successive approximation vector quantization. The usefulness of the proposed method is demonstrated by simulation. Using those results, the sampling process is reformulated as the vector quantization in the coordinate space. Thus, a new development is made for the variable rate sampling method.

International audience

There are a large number of methods in Content Based Image Retrieval (CBIR) systems that are based on color. The Global Color Histogram (GCH) is the most widely used method, the success of which flows from its computational simplicity. The problem with this method, however, is the absence of spatial information on the distribution of the color. To incorporate spatial information, Local Color Histograms (LCHs) are proposed, where, images are decomposed into several blocks. To reduce the space overhead of LCHs, Cell/Color histograms (CCHs) are proposed. In CCH, a cell histogram is calculated for each of the colors present in images. In this paper, we propose a new color based approach called Cell Color Coherence Vector (Cell-CCV). The proposed method is a fusion of two efficient color techniques: Cell Color Histogram (CCH) and Color Coherence Vector (CCV). Indeed, it uses a color coherence vector for each of the colors present in the images. CCV classifies pixels into coherent or incoherent regions. This gives good distinctions that cannot be made with color histograms used in CCH. Therefore, Cell-CCV takes advantage of both methods. The proposed method was tested using Wang dataset (Corel 1K). Results show that the newly developed method (Cell-CCV) performs significantly better than CCV and CCH for image retrieval in terms of weighted precision and average precision of 1000 queries.

End-to-end image and video codecs are becoming increasingly competitive, compared to traditional compression techniques that have been developed through decades of manual engineering efforts. These trainable codecs have many advantages over traditional techniques, such as their straightforward adaptation to perceptual distortion metrics and high performance in specific fields thanks to their learning ability. However, current state-of-the-art neural codecs do not fully exploit the benefits of vector quantization and the existence of the entropy gradient in decoding devices. In this paper, we propose to leverage these two properties (vector quantization and entropy gradient) to improve the performance of off-the-shelf codecs. Firstly, we demonstrate that using non-uniform scalar quantization cannot improve performance over uniform quantization. We thus suggest using predefined optimal uniform vector quantization to improve performance. Secondly, we show that the entropy gradient, available at the decoder, is correlated with the reconstruction error gradient, which is not available at the decoder. We therefore use the former as a proxy to enhance compression performance. Our experimental results show that these approaches save between 1 to 3% of the rate for the same quality across various pre-trained methods. In addition, the entropy gradient based solution improves traditional codec performance significantly as well.

In this paper, a color image retrieval scheme based on quadtree classified vector quantization (QCVQ) is proposed. This scheme not only captures intra-block correlation but also exploits the visual importance of image blocks to efficiently describe the content of images in a compressed domain. In the proposed algorithm, a query image is first divided by quadtree segmentation and then classified into smooth and high-detail blocks. For high-detail blocks, the local thresholding classifier with 28 edge binary templates is employed to extract a variety of visually important regions which are edge intensive. After all of the blocks in the image are encoded by the pre-trained QCVQ codebook, the indices in the compressed domain are obtained. Finally, the frequencies of indices are counted to build the index histogram as a feature of the query image. Simulation results demonstrate that our proposed scheme yields the better retrieval performance compared to the well-known vector quantization (VQ)-based image retrieval method and three other techniques. These results show that quadtree segmentation and edge style classification are indeed helpful for improving the performance of content-based image retrieval.

A vector quantizer maps a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</tex> -dimensional vector into one of a finite set of output vectors or "points". Although certain lattices have been shown to have desirable properties for vector quantization applications, there are as yet no algorithms available in the quantization literature for building quantizers based on these lattices. An algorithm for designing vector quantizers based on the root lattices <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A_{n}, D_{n}</tex> , and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">E_{n}</tex> and their duals is presented. Also, a coding scheme that has general applicability to all vector quantizers is presented. A four-dimensional uniform vector quantizer is used to encode Laplacian and gamma-distributed sources at entropy rates of one and two bits/sample and is demonstrated to achieve performance that compares favorably with the rate distortion bound and other scalar and vector quantizers. Finally, an application using uniform four- and eight-dimensional vector quantizers for encoding the discrete cosine transform coefficients of an image at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0.5</tex> bit/pel is presented, which visibly illustrates the performance advantage of vector quantization over scalar quantization.

In this correspondence, a modified version of Hunt's (1980) image model is used to interpret the distribution of image data vectors. The model suggests that the diagonal line of the coordinates system is a good approximation of the principal axis of the image data vector set. The validity of the model is supported by experiments. Following this suggestion, an effective nearest neighbor search method for vector quantization of image data is developed. The method is based on partitioning the vector space using hyperplanes which are perpendicular to the diagonal direction of the coordinate system. The validity of the method is assessed by analyzing its complexity and comparing its performance to those of existing algorithms on a number of images.