Image Representation Method Research Articles

1P1-U04 Informative Census Transform for Visual Object Representation(Robot Vision)

Our paper newly presents unsupervised feature representation method for images called informative census transform (ICT) based on statistical analysis of CT binary features and submodular optimization. A new cost function is designed to measure the informativeness of each binary feature by considering (1) an individual informativeness of features and (2) relative informativeness between binary features. Moreover, two constraints are designed by considering sub-modular characteristics to guarantee theoretical performance and fast optimization via simple greedy algorithm. Experimental results show that the proposed ICT features with two constraints outperforms the traditional CT features in terms of recognition performance and computational cost.

Augmenting multi-instance multilabel learning with sparse bayesian models for skin biopsy image analysis.

Skin biopsy images can reveal causes and severity of many skin diseases, which is a significant complement for skin surface inspection. Automatic annotation of skin biopsy image is an important problem for increasing efficiency and reducing the subjectiveness in diagnosis. However it is challenging particularly when there exists indirect relationship between annotation terms and local regions of a biopsy image, as well as local structures with different textures. In this paper, a novel method based on a recent proposed machine learning model, named multi-instance multilabel (MIML), is proposed to model the potential knowledge and experience of doctors on skin biopsy image annotation. We first show that the problem of skin biopsy image annotation can naturally be expressed as a MIML problem and then propose an image representation method that can capture both region structure and texture features, and a sparse Bayesian MIML algorithm which can produce probabilities indicating the confidence of annotation. The proposed algorithm framework is evaluated on a real clinical dataset containing 12,700 skin biopsy images. The results show that it is effective and prominent.

Research on Vocabulary Sizes and Codebook Universality

Codebook is an effective image representation method. By clustering in local image descriptors, a codebook is shown to be a distinctive image feature and widely applied in object classification. In almost all existing works on codebooks, the building of the visual vocabulary follows a basic routine, that is, extracting local image descriptors and clustering with a user-designated number of clusters. The problem with this routine lies in that building a codebook for each single dataset is not efficient. In order to deal with this problem, we investigate the influence of vocabulary sizes on classification performance and vocabulary universality with the kNN classifier. Experimental results indicate that, under the condition that the vocabulary size is large enough, the vocabularies built from different datasets are exchangeable and universal.

Discriminant analysis with Gabor phase feature for robust face recognition

An occlusion robust image representation method is presented and applied to face recognition. In our method, Gabor phase difference representation is used mainly to resist occlusion. Based on the good ability of Gabor filters to capture image structure and the robustness to image occlusion shown here, Gabor phase features are expected to be discriminative and robust for face representation in occlusion case. Furthermore, we find that different scales and orientations of Gabor phase features lead to quite varied performance and then we analyze it carefully and find the effective Gabor phase (EGP) features. Moreover, we adopt spectral regression–based discriminant analysis, along with the extracted EGP features, to find the most discriminant subspace for classification. Thereby, an occlusion robust face image discriminant subspace is derived. Five kinds of feature representation methods and two subspace learning methods are compared for our recognition problem. Extensive experiments with various occlusion cases show the efficacy of the proposed method.

Practical implementation of tetrahedral mesh reconstruction in emission tomography

This paper presents a practical implementation of image reconstruction on tetrahedral meshes optimized for emission computed tomography with parallel beam geometry. Tetrahedral mesh built on a point cloud is a convenient image representation method, intrinsically three-dimensional and with a multi-level resolution property. Image intensities are defined at the mesh nodes and linearly interpolated inside each tetrahedron. For the given mesh geometry, the intensities can be computed directly from tomographic projections using iterative reconstruction algorithms with a system matrix calculated using an exact analytical formula. The mesh geometry is optimized for a specific patient using a two stage process. First, a noisy image is reconstructed on a finely-spaced uniform cloud. Then, the geometry of the representation is adaptively transformed through boundary-preserving node motion and elimination. Nodes are removed in constant intensity regions, merged along the boundaries, and moved in the direction of the mean local intensity gradient in order to provide higher node density in the boundary regions. Attenuation correction and detector geometric response are included in the system matrix. Once the mesh geometry is optimized, it is used to generate the final system matrix for ML-EM reconstruction of node intensities and for visualization of the reconstructed images. In dynamic PET or SPECT imaging, the system matrix generation procedure is performed using a quasi-static sinogram, generated by summing projection data from multiple time frames. This system matrix is then used to reconstruct the individual time frame projections. Performance of the new method is evaluated by reconstructing simulated projections of the NCAT phantom and the method is then applied to dynamic SPECT phantom and patient studies and to a dynamic microPET rat study. Tetrahedral mesh-based images are compared to the standard voxel-based reconstruction for both high and low signal-to-noise ratio projection datasets. The results demonstrate that the reconstructed images represented as tetrahedral meshes based on point clouds offer image quality comparable to that achievable using a standard voxel grid while allowing substantial reduction in the number of unknown intensities to be reconstructed and reducing the noise.

Medical image fusion using multi-level local extrema

The fusion of data for medical imaging has become a central issue in such biomedical applications as image-guided surgery and radiotherapy. The multi-level local extrema (MLE) representation has been shown to have many advantages over conventional image representation methods. In this paper, we propose a new fusion algorithm for multi-modal medical images based on MLE. Our method enables the decomposition of input images into coarse and detailed layers in the MLE schema, and utilizes local energy and contrast fusion rules for coefficient selection in the different layers. This preserves more detail in the source images and further improves the quality of the fused image. The final fused image is obtained from the superposition of selected coefficients in the coarse and detailed layers. We illustrate the performance of the proposed method using three groups of medical images from different sources as our experimental subjects. We also compare our method with other techniques using cumulative mutual information, the objective image fusion performance measure, spatial frequency, and a blind quality index. Experimental results show that our method achieves a superior performance in both subjective and objective assessment criteria.

A Study on Image Representation Method Based on Biological Visual Mechanism

Image representation is a key issue among many image processing tasks. By considering the problems faced by current general image representation methods, such as excessive computing amount, sensitivity to noise, lack of self-adaptability etc, a novel image representation method based on biologic visual mechanisms is proposed in this paper. Through simulating the primary visual cortex to realize the sparse representation of outside image it also introduced the synchronization mechanism to make it more accordant with visual system. Finally the presented method was verified by applying it to compress natural images and digital literature images respectively. The result showed that this new representation method is better than the general sparse representation method on both aspects of compression ratio and noise sensitivity.

Object-based image retrieval with kernel on adjacency matrix and local combined features

In object-based image retrieval, there are two important issues: an effective image representation method for representing image content and an effective image classification method for processing user feedback to find more images containing the user-desired object categories. In the image representation method, the local-based representation is the best selection for object-based image retrieval. As a kernel-based classification method, Support Vector Machine (SVM) has shown impressive performance on image classification. But SVM cannot work on the local-based representation unless there is an appropriate kernel. To address this problem, some representative kernels are proposed in literatures. However, these kernels cannot work effectively in object-based image retrieval due to ignoring the spatial context and the combination of local features. In this article, we present Adjacent Matrix (AM) and the Local Combined Features (LCF) to incorporate the spatial context and the combination of local features into the kernel. We propose the AM-LCF feature vector to represent image content and the AM-LCF kernel to measure the similarities between AM-LCF feature vectors. According to the detailed analysis, we show that the proposed kernel can overcome the deficiencies of existing kernels. Moreover, we evaluate the proposed kernel through experiments of object-based image retrieval on two public image sets. The experimental results show that the performance of object-based image retrieval can be improved by the proposed kernel.

Kernel Density Feature Points Estimator for Content-Based Image Retrieval

Research is taking place to find effective algorithms for content-based image representation and description. There is a substantial amount of algorithms available that use visual features (color, shape, texture). Shape feature has attracted much attention from researchers that there are many shape representation and description algorithms in literature. These shape image representation and description algorithms are usually not application independent or robust, making them undesirable for generic shape description. This paper presents an object shape representation using Kernel Density Feature Points Estimator (KDFPE). In this method, the density of feature points within defined rings around the centroid of the image is obtained. The KDFPE is then applied to the vector of the image. KDFPE is invariant to translation, scale and rotation. This method of image representation shows improved retrieval rate when compared to Density Histogram Feature Points (DHFP) method. Analytic analysis is done to justify our method, which was compared with the DHFP to prove its robustness.

A New Image Representation Method Using Nonoverlapping Non-symmetry and Anti-packing Model for Medical Images

Inspired by the idea of the S-Tree Coding (STC) and by extending the popular Gouraud shading approach, in this paper, we propose a new image representation method using the Nonoverlapping Non-symmetry and Anti-packing Model (NNAM) for medical images, which is called the NNAM method. Also, a raster-first strategy for searching a rectangle subpattern in the NNAM method is put forward. During the procedure of scanning a rectangle subpptern, the value of the horizontal ordinate is firstly increased, and then the value of the vertical ordinate is increased until the rectangle subpattern finally becomes a non-homogeneous block. By comparing our proposed NNAM method with the conventional STC method, the experimental results presented in this paper show that the former can significantly reduce the lower bit rate and the number of homogenous blocks than the latter whereas remaining the satisfactory image quality. Also, our proposed NNAM method for medical images, as envisaged in this paper, shows a very strong promise and it has good potential in business applications dealing with image processing, such as reducing storage room, increasing processing speed, and improving pattern match efficiency.

A Square NAM Representation Method for Binary Images

The Non-symmetry and Anti-packing Model (NAM) is an effective pattern representation model. In this paper, a new NAM image representation method by using the nonoverlapping square subpatterns, which is called the square NAM (SNAM) representation, is proposed. The idea of the SNAM representation is described. The square subpattern has its own significant characteristics when it is compared with other subpatterns such as the triangle and the rectangle, etc. Unlike the general triangle subpattern which needs to record three vertices of the triangle and unlike the rectangle subpattern which needs to record the coordinates of its starting point, length and width, the square subpattern needs to record only its starting point and the side. Therefore, as far as a single record of the square subpattern is concerned, it can save storage space more effectively. The theoretical and experimental results presented in this paper prove the efficiency and the effectiveness of the proposed SNAM representation method for binary images.

A Study on Method of Visual Image Representation in Decentralized Space

A Study on Method of Visual Image Representation in Decentralized Space

A Novel Gray Image Representation Method Based on NAM Using Nonoverlapping Square Subpatterns

In this paper, we propose a novel gray image representation method based on the non-symmetry and anti-packing model (NAM) by using the nonoverlapping square subpatterns, which is called the square NAM for gray images (SNAMG) representation method. Also, a SNAMG representation algorithm is put forward and the storage structures, the total data amount, and the time complexity of the proposed algorithm are analyzed in detail. By taking some standard gray images, such as ‘F16’ and ‘Peppers’, as the typical test objects, and comparing the proposed algorithm with those of the triangle NAM for gray images (TNAMG) and the classic linear quadtree (LQT), the theoretical and experimental results show that the former is obviously superior to the latter with respect to the numbers of subpatterns (nodes) and the data storage, and therefore it is a better method to represent the gray image pattern.

An Improved RNAMC Image Representation

In this paper, by controlling the ratio of the length and the width of a homogenous block, we proposed an improved algorithm for the gray image representation by using the Rectangular Non-symmetry and Anti-packing Model Coding (RNAMC) and extended shading approach, which is called the IRNAMC image representation method. Also, we present an IRNAMC representation algorithm of gray images. By comparing our proposed IRNAMC method with the conventional S-Tree Coding (STC) method, the experimental results presented in this paper show that the former can significantly reduce the lower bit rate and the number of homogenous blocks than the latter whereas remaining the satisfactory image quality. Also, the experimental results show that by controlling the ratio of the length and the width, we can improve the reconstructed image quality of the RNAMC method.

Combining wavelets transform and Hu moments with self-organizing maps for medical image categorization

Images are fundamental sources of information in modern medicine. The images stored in a database and divided in categories are an important step for image retrieval. For an automatic categorization process, detailed analysis is done regarding image representation and generalization method. The baseline method for this process, in the medical image context, is using thumbnails and K-nearest neighbor (KNN), which is easily implemented and has had satisfactory results in literature. This work addresses an alternative method for automatic categorization, which jointly uses discrete wavelet transform with Hu's moments for image representation and self-organizing maps (SOM) neural networks combined with the KNN classifier (SOM-KNN), for medical image categorization. Furthermore, extensive experiments are conducted, to define the best wavelet family and to select the best coefficients set, to consider the remaining wavelet coefficients set (not selected as the best ones) through their Hu's moments, and to carry out a contrastive study with other successful approaches for categorization. The categorization result from a database with 10,000 images in 116 categories yielded 81.8% of correct rate, which is much better than the 67.9% obtained by the baseline method; and the time consumed in classification processing with SOM-KNN is 100 times shorter than KNN.

A new non-symmetry and anti-packing model and its application to image contrast enhancement

In this paper, a non-symmetry and anti-packing image representation model (NAM) has been proposed. NAM is a hierarchical image representation method and it aims to provide faster operations and less storage requirement. By taking a rectangle sub-pattern, for example, we describe the idea of NAM and its encoding algorithm. In addition, an approach for adaptive area histogram equalization for image contrast enhancement based on a NAM image is presented. The contrast enhancement approach is designed to meet the NAM image representation and it can be duplicated with faster operation. The complexity analysis and the experimental results show that the NAM based algorithm for image contrast enhancement is faster and more effective than that based on matrix image.

Image representation for generic object recognition using higher-order local autocorrelation features on posterior probability images

This paper presents a novel image representation method for generic object recognition by using higher-order local autocorrelations on posterior probability images. The proposed method is an extension of the bag-of-features approach to posterior probability images. The standard bag-of-features approach is approximately thought of as a method that classifies an image to a category whose sum of posterior probabilities on a posterior probability image is maximum. However, by using local autocorrelations of posterior probability images, the proposed method extracts richer information than the standard bag-of-features. Experimental results reveal that the proposed method exhibits higher classification performances than the standard bag-of-features method.

Bag-of-Visual-Words Based on Clonal Selection Algorithm for SAR Image Classification

Synthetic aperture radar (SAR) image classification involves two crucial issues: suitable feature representation technique and effective pattern classification methodology. Here, we concentrate on the first issue. By exploiting a famous image feature processing strategy, Bag-of-Visual-Words (BOV) in image semantic analysis and the artificial immune systems (AIS)'s abilities of learning and adaptability to solve complicated problems, we present a novel and effective image representation method for SAR image classification. In BOV, an effective fused feature sets for local feature representation are first formulated, which are viewed as the low-level features in it. After that, clonal selection algorithm (CSA) in AIS is introduced to optimize the prediction error of k-fold cross-validation for getting more suitable visual words from the low-level features. Finally, the BOV features are represented by the learned visual words for subsequent pattern classification. Compared with the other four algorithms, the proposed algorithm obtains more satisfactory and cogent classification experimental results.

Improved rectangle NAM algorithm for image representation

In order to improve image representation efficiency,an improved rectangle Non-symmetry Anti-packing representation Model(NAM) named IRNAM was proposed for image coding.This scheme adopted double-rectangle sub-patterns to represent gray image,integrated bit plane optimization strategy and stored the sub-patterns continuously,thus the number of sub-patterns had decreased sharply.The experimental results show that compared with rectangle NAM algorithm and other improved NAM algorithms,the number of sub-patterns can be reduced apparently after employing IRNAM algorithm,which can save data storage space effectively and prove to be a high performance image representation method.

Image representation using Laplacian regularized nonnegative tensor factorization

Tensor provides a better representation for image space by avoiding information loss in vectorization. Nonnegative tensor factorization (NTF), whose objective is to express an n-way tensor as a sum of k rank-1 tensors under nonnegative constraints, has recently attracted a lot of attentions for its efficient and meaningful representation. However, NTF only sees Euclidean structures in data space and is not optimized for image representation as image space is believed to be a sub-manifold embedded in high-dimensional ambient space. To avoid the limitation of NTF, we propose a novel Laplacian regularized nonnegative tensor factorization (LRNTF) method for image representation and clustering in this paper. In LRNTF, the image space is represented as a 3-way tensor and we explicitly consider the manifold structure of the image space in factorization. That is, two data points that are close to each other in the intrinsic geometry of image space shall also be close to each other under the factorized basis. To evaluate the performance of LRNTF in image representation and clustering, we compare our algorithm with NMF, NTF, NCut and GNMF methods on three standard image databases. Experimental results demonstrate that LRNTF achieves better image clustering performance, while being more insensitive to noise.