Learning bayesian classifiers for scene classification with a visual grammar

A challenging problem in image content extraction and classification is building a system that automatically learns high-level semantic interpretations of images. We describe a Bayesian framework for a visual grammar that aims to reduce the gap between low-level features and high-level user semantics. Our approach includes modeling image pixels using automatic fusion of their spectral, textural, and other ancillary attributes; segmentation of image regions using an iterative split-and-merge algorithm; and representing scenes by decomposing them into prototype regions and modeling the interactions between these regions in terms of their spatial relationships. Naive Bayes classifiers are used in the learning of models for region segmentation and classification using positive and negative examples for user-defined semantic land cover labels. The system also automatically learns representative region groups that can distinguish different scenes and builds visual grammar models. Experiments using Landsat scenes show that the visual grammar enables creation of high-level classes that cannot be modeled by individual pixels or regions. Furthermore, learning of the classifiers requires only a few training examples.     

A complexity reduction technique for image vector quantization

A technique for reducing the complexity of spatial-domain image vector quantization (VQ) is proposed. The conventional spatial domain distortion measure is replaced by a transform domain subspace distortion measure. Due to the energy compaction properties of image transforms, the dimensionality of the subspace distortion measure can be reduced drastically without significantly affecting the performance of the new quantizer. A modified LBG algorithm incorporating the new distortion measure is proposed. Unlike conventional transform domain VQ, the codevector dimension is not reduced and a better image quality is guaranteed. The performance and design considerations of a real-time image encoder using the techniques are investigated. Compared with spatial domain a speed up in both codebook design time and search time is obtained for mean residual VQ, and the size of fast RAM is reduced by a factor of four. Degradation of image quality is less than 0.4 dB in PSNR.     

Reversible data embedding for high quality images using interpolation and reference pixel distribution mechanism

This paper proposes a reversible data hiding method based on image interpolation and the detection of smooth and complex regions in the cover images. A binary image that represents the locations of reference pixels is constructed according the local image activity. In complex regions, more reference pixels are chosen and, thus, fewer pixels are used for embedding, which reduces the image degradation. On the other hand, in smooth regions, less reference pixels are chosen, which increases the embedding capacity without introducing significant distortion. Pixels are interpolated according to the constructed binary image, and the interpolation errors are then used to embed data through histogram shifting. The pixel values in the cover image are modified one grayscale unit at most to ensure that a high quality stego image can be produced. The experimental results show that the proposed method provides better image quality and embedding capacity compared with prior works.     

Weighted universal image compression

We describe a general coding strategy leading to a family of universal image compression systems designed to give good performance in applications where the statistics of the source to be compressed are not available at design time or vary over time or space. The basic approach considered uses a two-stage structure in which the single source code of traditional image compression systems is replaced with a family of codes designed to cover a large class of possible sources. To illustrate this approach, we consider the optimal design and use of two-stage codes containing collections of vector quantizers (weighted universal vector quantization), bit allocations for JPEG-style coding (weighted universal bit allocation), and transform codes (weighted universal transform coding). Further, we demonstrate the benefits to be gained from the inclusion of perceptual distortion measures and optimal parsing. The strategy yields two-stage codes that significantly outperform their single-stage predecessors. On a sequence of medical images, weighted universal vector quantization outperforms entropy coded vector quantization by over 9 dB. On the same data sequence, weighted universal bit allocation outperforms a JPEG-style code by over 2.5 dB. On a collection of mixed test and image data, weighted universal transform coding outperforms a single, data-optimized transform code (which gives performance almost identical to that of JPEG) by over 6 dB.     

Embedded image compression based on wavelet pixel classification and sorting

The method of modeling and ordering in wavelet domain is very important to design a successful algorithm of embedded image compression. In this paper, the modeling is limited to "pixel classification," the relationship between wavelet pixels in significance coding. Similarly, the ordering is limited to "pixel sorting," the coding order of wavelet pixels. We use pixel classification and sorting to provide a better understanding of previous works. The image pixels in wavelet domain are classified and sorted, either explicitly or implicitly, for embedded image compression. A new embedded image code is proposed based on a novel pixel classification and sorting (PCAS) scheme in wavelet domain. In PCAS, pixels to be coded are classified into several quantized contexts based on a large context template and sorted based on their estimated significance probabilities. The purpose of pixel classification is to exploit the intraband correlation in wavelet domain. Pixel sorting employs several fractional bit-plane coding passes to improve the rate-distortion performance. The proposed pixel classification and sorting technique is simple, yet effective, producing an embedded image code with excellent compression performance. In addition, our algorithm is able to provide either spatial or quality scalability with flexible complexity.     

No-reference image quality assessment based on non-subsample shearlet transform and natural scene statistics

A novel no-reference (NR) image quality assessment (IQA) method is proposed for assessing image quality across multifarious distortion categories. The new method transforms distorted images into the shearlet domain using a non-subsample shearlet transform (NSST), and designs the image quality feature vector to describe images utilizing natural scenes statistical features:coefficient distribution, energy distribution and structural correlation (SC) across orientations and scales. The final image quality is achieved from distortion classification and regression models trained by a support vector machine (SVM). The experimental results on the LIVE2 IQA database indicate that the method can assess image quality effectively, and the extracted features are susceptive to the category and severity of distortion. Furthermore, our proposed method is database independent and has a higher correlation rate and lower root mean squared error (RMSE) with human perception than other high performance NR IQA methods.     

Block classification for an adaptive 1-D/2-D DCT video coding

A classification scheme for an adaptive one- or two-dimensional discrete cosine transform (1-D/2-D DCT) technique is described and demonstrated to be a more appropriate strategy than the conventional 2-D DCT for coding motion compensated prediction error images. Two block-based classification methods are introduced and their accuracy in predicting the correct transform type discussed. The accuracy is assessed with a classification measure designed to ascertain the effectiveness of energy compaction when the predicted transform class is applied; vis-a-vis horizontally, vertically or two-dimensionally transformed blocks. Energy compaction is a useful property not only for efficient entropy coding but also for enhancing the resilience of the transform coder to quantisation noise. Improvements against the homogeneous 2-D DCT system both in terms of the peak signal to noise ratio and subjective assessments are achieved. Observable ringing artifacts along edges, which are usual in conventional transform coding, are reduced     

Adaptive image coding using spectral classification

We present a new classification scheme, dubbed spectral classification, which uses the spectral characteristics of the image blocks to classify them into one of a finite number of classes. A vector quantizer with an appropriate distortion measure is designed to perform the classification operation. The application of the proposed spectral classification scheme is then demonstrated in the context of adaptive image coding. It is shown that the spectral classifier outperforms gain-based classifiers while requiring a lower computational complexity.     

Progressive Image Transmission Using LOT with Two-Channel Conjugate VQ

A progressive image transmission (PIT) scheme based on the classified two-channel conjugate VQ (TCCVQ) technique in the lapped orthogonal transform (LOT) domain is proposed. Conventional block transform coding of images using DCT produces in general undesirable block-artifacts at low bit rates. In this paper, image blocks are transformed using the LOT and classified into four classes based on their structural properties and further subdivided adaptively into subvectors depending on the LOT coefficient statistics to improve the reconstructed image quality by adaptive bit allocation. The subvectors are quantized using the two-channel conjugate VQ technique which has less computational complexity and less storage memory, and is more robust against channel errors. The vector quantized subvectors are transmitted and decoded in stages so that an image is progressively reconstructed, i.e., initially a crude version followed by quality build up in successive stages as the occasion demands in interactive visual communications. Coding tests using computer simulations show that the LOT/TCCVQ-based PIT of images is an effective coding scheme. The results are also compared with those obtained using conventional classified VQ in both the DCT and LOT domains.     

Classified vector quantisation with variable block-size DCT models

The paper describes the classified vector quantisation (CVQ) of an image, based on quadtrees and a classification technique in the discrete cosine transform (DCT) domain. In this scheme, a quadtree is used to segment low-detail regions into variable sized blocks and high-detail regions into uniform 4×4 blocks of various edge and mixed classes. High-detail blocks are classified by an edge-oriented classifier which employs a pattern-matching technique with edge models defined in the normalised DCT domain. The proposed classifier is simple to implement, and efficiently classifies edges to good visual accuracy. The low-detail regions are encoded at very low bit rates with little perceptual degradation, while the encoding of the high-detail regions is performed to achieve a good perceptual quality in the decoded image. Decoded images of high visual quality are obtained for encoding rates between 0.3 and 0.7 bpp     

Image Fusion Processing for IKONOS 1-m Color Imagery

Many image fusion techniques have been developed. However, most existing fusion processes produce color distortion in 1-m fused IKONOS images due to nonsymmetrical spectral responses of IKONOS imagery. Here, we proposed a fusion process to minimize this spectral distortion in IKONOS 1-m color images. The 1-m fused image is produced from a 4-m multispectral (MS) and 1-m panchromatic (PAN) image, maintaining the relations of spectral responses between PAN and each band of the MS images. To obtain this relation, four spectral weighting parameters are added with the pixel value of each band of the original MS image. Then, each pixel value is updated using a steepest descent method to reflect the maximum spectral response on the fused image. Comparison among the proposed technique and existing processes [intensity hue saturation (IHS) image fusion, Brovey transform, principal component analysis, fast IHS image fusion] has been done. Our proposed technique has succeeded to generate 1-m fused images where spectral distortion has been reduced significantly, although some block distortions appeared at the edge of the fused images. To remove this block distortion, we also proposed a sharpening process using a wavelet transform, which removed block distortion without significant change in the color of the entire image.     

Compression of multispectral images by three-dimensional SPIHTalgorithm

The authors carry out low bit-rate compression of multispectral images by means of the Said and Pearlman's SPIHT algorithm, suitably modified to take into account the interband dependencies. Two techniques are proposed: in the first, a three-dimensional (3D) transform is taken (wavelet in the spatial domain, Karhunen-Loeve in the spectral domain) and a simple 3D SPIHT is used; in the second, after taking a spatial wavelet transform, spectral vectors of pixels are vector quantized and a gain-driven SPIHT is used. Numerous experiments on two sample multispectral images show very good performance for both algorithms     

Automatic Spectral Rule-Based Preliminary Mapping of Calibrated Landsat TM and ETM+ Images

Based on purely spectral-domain prior knowledge taken from the remote sensing (RS) literature, an original spectral (fuzzy) rule-based per-pixel classifier is proposed. Requiring no training and supervision to run, the proposed spectral rule-based system is suitable for the preliminary classification (primal sketch, in the Marr sense) of Landsat-5 Thematic Mapper and Landsat-7 Enhanced Thematic Mapper Plus images calibrated into planetary reflectance (albedo) and at-satellite temperature. The classification system consists of a modular hierarchical top-down processing structure, which is adaptive to image statistics, computationally efficient, and easy to modify, augment, or scale to other sensors' spectral properties, like those of the Advanced Spaceborne Thermal Emission and Reflection Radiometer and of the Satellite Pour l'Observation de la Terre (SPOT-4 and -5). As output, the proposed system detects a set of meaningful and reliable fuzzy spectral layers (strata) consistent (in terms of one-to-one or many-to-one relationships) with land cover classes found in levels I and II of the U.S. Geological Survey classification scheme. Although kernel spectral categories (e.g., strong vegetation) are detected without requiring any reference sample, their symbolic meaning is intermediate between those (low) of clusters and segments and those (high) of land cover classes (e.g., forest). This means that the application domain of the kernel spectral strata is by no means alternative to RS data clustering, image segmentation, and land cover classification. Rather, prior knowledge-based kernel spectral categories are naturally suitable for driving stratified application-specific classification, clustering, or segmentation of RS imagery that could involve training and supervision. The efficacy and robustness of the proposed rule-based system are tested in two operational RS image classification problems.     

Neural-fuzzy classification for segmentation of remotely sensedimages

An unsupervised classification technique conceptualized in terms of neural and fuzzy disciplines for the segmentation of remotely sensed images is presented. The process consists of three major steps: 1) pattern transformation; 2) neural classification; 3) fuzzy grouping. In the first step, the multispectral patterns of image pixels are transformed into what we call coarse patterns. In the second step, a delicate classification of pixels is attained by applying an ART neural classifier to the transformed pixel patterns. Since the resultant clusters of pixels are usually too keen to be of practical significance, in the third step, a fuzzy clustering algorithm is invoked to integrate pixel clusters. A function for measuring clustering validity is defined with which the optimal number of classes can be automatically determined by the clustering algorithm. The proposed technique is applied to both synthetic and real images. High classification rates have been achieved for synthetic images. We also feel comfortable with the results of the real images because their spectral variances are even smaller than the spectral variances of the synthetic images examined     

基于分类和3D-SPIHT算法相结合的多光谱图像压缩

文中提出了一种基于分类预测的三维SPIHT算法,并对多光谱1~7波段图像进行了压缩实验。首先对图像数据作三维变换,空域采用浮点97小波去除相关性,谱域分类预测去除冗余;再根据分类预测算法获得系数的残差图像,并对残差图像进行三维SPIHT编码; 而对分类预测时得到的码书和索引表进行哈夫曼无损压缩; 将这3个编码文件传送到解码端用于图像重构。实验证明该算法具有很好的重构效果。     

A lossless compression scheme for Bayer color filter array images.

In most digital cameras, Bayer color filter array (CFA) images are captured and demosaicing is generally carried out before compression. Recently, it was found that compression-first schemes outperform the conventional demosaicing-first schemes in terms of output image quality. An efficient prediction-based lossless compression scheme for Bayer CFA images is proposed in this paper. It exploits a context matching technique to rank the neighboring pixels when predicting a pixel, an adaptive color difference estimation scheme to remove the color spectral redundancy when handling red and blue samples, and an adaptive codeword generation technique to adjust the divisor of Rice code for encoding the prediction residues. Simulation results show that the proposed compression scheme can achieve a better compression performance than conventional lossless CFA image coding schemes.     

基于二次反锐化掩模的多特征侧扫 声纳成像海底底质分类

提出了一种二次反锐化掩模算子(QUM),用于从图像中提取等比重的边缘和细节信息.把图像的平均边缘细节量(QUM )、平均纹理能量(TEM )和标准差(σ)相结合,构成本文的三维特征矢量(QUM ,TEM ,σ),用于对侧扫声纳海底图像进行底质分类.利用该特征矢量(QUM ,TEM ,σ)对泥、沙、石三种类型海底的150幅侧扫声纳图像进行分类实验,获得了最高96.7%、最低90.7%的识别率,而利用常用的灰阶共生矩阵方法的分类识别率为87.3%,表明本文方法能较好地用于海底底质分类.     

Still image coding based on vector quantization and fractalapproximation

In this paper, we propose a coding algorithm for still images using vector quantization (VQ) and fractal approximation, in which low-frequency components of an input image are approximated by VQ, and its residual is coded by fractal mapping. The conventional fractal coding algorithms indirectly used the gray patterns of an original image with contraction mapping, whereas the proposed fractal coding method employs an approximated and then decimated image as a domain pool and uses its gray patterns. Thus, the proposed algorithm utilizes fractal approximation without the constraint of contraction mapping. For approximation of an original image, we employ the discrete cosine transform (DCT) rather than conventional polynomial-based transforms. In addition, for variable blocksize segmentation, we use the fractal dimension of a block that represents the roughness of the gray surface of a region. Computer simulations with several test images show that the proposed method shows better performance than the conventional fractal coding methods for encoding still pictures.     

基于分类的高光谱图像压缩算法

高光谱图像庞大的数据量给存储与传输带来巨大挑战,必须采用有效的压缩算法对其进行压缩。提出了一种基于分类的高光谱图像有损压缩算法。首先利用C均值算法对高光谱图像进行无监督光谱分类。根据分类图,针对每一类数据分别采用自适应KLT(Karhunen-Love transform)进行谱间去相关;然后对每个主成分分别进行二维小波变换。为了获得最佳的率失真性能,采用EBCOT(Embedded Block Coding with Optimized Truncation)算法对所有的主成分进行联合率失真编码。实验结果表明,所提出算法的有损压缩性能优于其它经典的压缩算法。     

Improved decoder for transform coding with application to the JPEGbaseline system

Transform coding, a simple yet efficient image coding technique, has been adopted by the Joint Photographic Experts Group (JPEG) as the basis for an emerging coding standard for compression of still images. However, for any given transform encoder, the conventional inverse transform decoder is suboptimal. Better performance can be obtained by a nonlinear interpolative decoder that performs table lookups to reconstruct the image blocks from the code indexes. Each received code index of an image block addresses a particular codebook to fetch a component vector. The image block can be reconstructed as the sum of the component vectors for that block. An iterative algorithm for designing a set of locally optimal codebooks is developed. Computer simulation results demonstrate that this improved decoding technique can be applied in the JPEG baseline system to decode enhanced quality pictures from the bit stream generated by the standard encoding scheme