Chong-Sze Tong

Texture Classification Using Refined Histogram

In this correspondence, we propose a novel, efficient, and effective Refined Histogram (RH) for modeling the wavelet subband detail coefficients and present a new image signature based on the RH model for supervised texture classification. Our RH makes use of a step function with exponentially increasing intervals to model the histogram of detail coefficients, and the concatenation of the RH model parameters for all wavelet subbands forms the so-called RH signature. To justify the usefulness of the RH signature, we discuss and investigate some of its statistical properties. These properties would clarify the sufficiency of the signature to characterize the wavelet subband information. In addition, we shall also present an efficient RH signature extraction algorithm based on the coefficient-counting technique, which helps to speed up the overall classification system performance. We apply the RH signature to texture classification using the well-known databases. Experimental results show that our proposed RH signature in conjunction with the use of symmetrized Kullback-Leibler divergence gives a satisfactory classification performance compared with the current state-of-the-art methods.

Statistical Wavelet Subband Characterization Based on Generalized Gamma Density and Its Application in Texture Retrieval

The modeling of image data by a general parametric family of statistical distributions plays an important role in many applications. In this paper, we propose to adopt the three-parameter generalized gamma density (G¿D) for modeling wavelet detail subband histograms and for texture image retrieval. The advantage of G¿D over the existing generalized Gaussian density (GGD) is that it provides more flexibility to control the shape of model which is critical for practical histogram-based applications. To measure the discrepancy between G¿Ds, we use the symmetrized Kullback-Leibler distance (SKLD) and derive a closed form for the SKLD between G¿Ds. Such a distance can be computed directly and effectively via the model parameters, making our proposed scheme particularly suitable for image retrieval systems with large image database. Experimental results on the well-known databases reveal the superior performance of our proposed method compared with the current existing approaches.

A Fast and Effective Model for Wavelet Subband Histograms and Its Application in Texture Image Retrieval

This paper presents a novel, effective, and efficient characterization of wavelet subbands by bit-plane extractions. Each bit plane is associated with a probability that represents the frequency of 1-bit occurrence, and the concatenation of all the bit-plane probabilities forms our new image signature. Such a signature can be extracted directly from the code-block code-stream, rather than from the de-quantized wavelet coefficients, making our method particularly adaptable for image retrieval in the compression domain such as JPEG2000 format images. Our signatures have smaller storage requirement and lower computational complexity, and yet, experimental results on texture image retrieval show that our proposed signatures are much more cost effective to current state-of-the-art methods including the generalized Gaussian density signatures and histogram signatures

Supervised Texture Classification Using Characteristic Generalized Gaussian Density

Abstract Generalized Gaussian density (GGD) is a well established model for high frequency wavelet subbands and has been applied in texture image retrieval with very good results. In this paper, we propose to adopt the GGD model in a supervised learning context for texture classification. Given a training set of GGDs, we define a characteristic GGD (CGGD) that minimizes its Kullback-Leibler distance (KLD) to the training set. We present mathematical analysis that proves the existence of our characteristic GGD and provide a sufficient condition for the uniqueness of CGGD, thus establishing a theoretical basis for its use. Our experimental results show that the proposed CGGD signature together with the use of KLD has a very promising recognition performance compared with existing approaches.

Statistical Properties of Bit-Plane Probability Model and Its Application in Supervised Texture Classification

The modeling of wavelet subband histograms via the product Bernoulli distributions (PBD) has received a lot of interest and the PBD model has been applied successfully in texture image retrieval. In order to fully understand the usefulness and effectiveness of the PBD model and its associated signature, namely, the bit-plane probability (BP) signature on image processing applications, we discuss and investigate some of their statistical properties. These properties would help to clarify the sufficiency of the BP signature to characterize wavelet subbands, which, in turn, justifies its use in real time applications. We apply the BP signature on supervised texture classification problem and experimental results suggest that the weighted L1-norm (rather than the standard L1-norm) should be used for the BP signature. Comparative classification experiments show that our method outperforms the current state-of-the-art Generalized Gaussian Density approaches.

Image Segmentation Using Fuzzy Region Competition and Spatial/Frequency Information

This paper presents a multiphase fuzzy region competition model that takes into account spatial and frequency information for image segmentation. In the proposed energy functional, each region is represented by a fuzzy membership function and a data fidelity term that measures the conformity of spatial and frequency data within each region to (generalized) Gaussian densities whose parameters are determined jointly with the segmentation process. Compared with the classical region competition model, our approach gives soft segmentation results via the fuzzy membership functions, and moreover, the use of frequency data provides additional region information that can improve the overall segmentation result. To efficiently solve the minimization of the energy functional, we adopt an alternate minimization procedure and make use of Chambolles fast duality projection algorithm. We apply the proposed method to synthetic and natural textures as well as real-world natural images. Experimental results show that our proposed method has very promising segmentation performance compared with the current state-of-the-art approaches.

Characterization of shapes for use in classification of starch grains images.

As tradition Chinese herbal medicine becomes increasingly popular, there is an urgent need for efficient and accurate methods for the authentication of the Chinese Materia Medica (CMM) used in the herbal medicine. In this work, we present a denoising filter and introduce the use of chord length distribution (CLD) for the classification of starch grains in microscopic images of Chinese Materia Medica. Our simple denoising filter is adaptive to the background and is shown to be effective to remove noise, which appears in CMM microscopic starch grains images. The CLD is extracted by considering the frequency of the chord length in the binarized starch grains image, and we shall show that the CLD is an efficient and effective characterization of the starch grains. Experimental results on 240 starch grains images of 24 classes show that our method outperforms benchmark result using the current state‐of‐the‐art method based on circular size distribution extracted by morphological operators at much higher computational cost. cost. Microsc. Res. Tech., 2008.

A novel and effective multistage classification system for microscopic starch grain images

This article presents a novel and effective multistage system for classifying Chinese Materia Medica microscopic starch grain images. The proposed classification system is constructed based on the Gaussian mixture model‐based clustering, the feature assignment algorithm, and the similarity measurement. Several features for each starch grain image are extracted and every class of drug is represented by a set of characteristic features. For each stage of the system, only one feature is chosen and assigned to that stage via the feature assignment algorithm, and the corresponding characteristic features are subdivided into smaller subsets based on clustering techniques. At the final stage, each subset contains a certain class of drugs (with corresponding characteristic features) and similarity measurement is carried out for starch grain classification. Three sets of the current state‐of‐the‐art starch grain features including the granulometric size distribution, the chord length distribution, and the wavelet signature are used to construct the system. Experimental results on a database of 240 images of 24 classes of drugs reveal the superior performance of the multistage system. Comparison with the traditional starch grain classification approaches indicates that our proposed multistage method produces a marked improvement in classification performance. Microsc. Res. Tech. 2009.

Supervised Learning Using Characteristic Generalized Gaussian Density and Its Application to Chinese Materia Medica Identification

This paper presents the estimation of the characteristic generalized Gaussian density (CGGD) given a set of known GGD distributions based on some optimization techniques, and its application to the Chinese Materia Medica identification. The CGGD parameters are estimated by minimizing the distance between the CGGD distribution and known GGD distributions. Our experimental results show that the proposed signature based on the CGGD together with the use of Kullback-Leibler distance outperforms the traditional wavelet-based energy signature. The recognition rate for the proposed method is higher than the energy signature by at least 10% to around 60% – 70%. Nevertheless, the extraction of CGGD estimators still retains comparable level of computational complexity. In general, our proposed method is very competitive compared with many other existing Chinese Materia Medica classification methods.

A Correlated Bit-Plane Model for Wavelet Subband Histograms and Its Application to Chinese Materia Medica Starch Grains Classification

This paper presents an effective statistical model for wavelet high frequency subband histograms and a novel image signature by bit-plane extractions. Our proposed model, namely, the first order correlated bit-plane probability model, is shown to match well with the observed histograms especially when the size of subband coefficients is small and performs better than the product Bernoulli distributions (PBD model) as described. Experimental results on supervised Chinese Materia Medica starch grains images classification show that our proposed signature based on wavelet subband correlated bit-plane probabilities outperforms the current state-of-the-art signatures including the generalized Gaussian density signature (GGD), the granulometric circular size distribution, and the bit-plane probability (BP) signature.