Jose Gerardo Rosiles

Texture classification with a biorthogonal directional filter bank

Classifying textures is a problem that has been considered by many researchers. Many of the high performance methods are based on extracting features from the textures and performing classification in the feature space. In this paper, we consider the application of a new directional filter bank (DFB) to the problem of texture classification. The DFB is used to provide a compact and efficient representation in which fast classification can be performed using classical statistical methods. The resulting method is shown to yield higher performance than feature-based techniques reported previously. Furthermore, the approach has the added attraction that both the computational complexity and storage requirements are relatively low. Experimental comparisons using the Brodatz texture database are also presented.

Image denoising using directional filter banks

The use of wavelet thresholding has been investigated with much success in the areas of denoising, density estimation, image restoration, etc. Significant attention has been given to wavelet thresholding as a signal denoising technique. The algorithm is simple and provides good results. The 2-D discrete wavelet transform (DWT) and its relatives have been used to generalize the denoising methods to images. However the DWT is limited in its representation of directional information like edges and some types of texture. We propose the use of a directional filter bank for image denoising under the same premise as wavelet thresholding: small magnitude subband coefficients represent noise and can be replaced with zeros while large coefficients reflect, in our case, strong signal content in a given direction. We show that the directional filter bank is capable of preserving edge information better than DWT based techniques while effectively removing noise. The proposed technique provides sharp images with higher perceptual quality.

A Low-Complexity Geometric Bilateration Method for Localization in Wireless Sensor Networks and Its Comparison with Least-Squares Methods

This research presents a distributed and formula-based bilateration algorithm that can be used to provide initial set of locations. In this scheme each node uses distance estimates to anchors to solve a set of circle-circle intersection (CCI) problems, solved through a purely geometric formulation. The resulting CCIs are processed to pick those that cluster together and then take the average to produce an initial node location. The algorithm is compared in terms of accuracy and computational complexity with a Least-Squares localization algorithm, based on the Levenberg–Marquardt methodology. Results in accuracy vs. computational performance show that the bilateration algorithm is competitive compared with well known optimized localization algorithms.

Automatic Dust Storm Detection Based on Supervised Classification of Multispectral Data

This paper address the detection of dust storms based on a probabilistic analysis of multispectral images. We develop a feature set based on the analysis of spectral bands reported in the literature. These studies have focused on the visual identification of the image channels that reflect the presence of dust storms through correlation with meteorological reports. Using this feature set we develop a Maximum Likelihood classifier and a Probabilistic Neural Network (PNN) to automate the dust storm detection process. The data sets are MODIS multispectral bands from NASA Terra satellite. Findings indicate that the PNN provides improved classification performance with reference to the ML classifier. Furthermore, the proposed schemes allow real-time processing of satellite data at 1 km resolutions which is an improvement compared to the 10 km resolution currently provided by other detection methods.

Directional analysis and filtering for dust storm detection in NOAA-AVHRR imagery

In this paper, we propose spatio-spectral processing techniques for the detection of dust storms and automatically finding its transport direction in 5-band NOAA-AVHRR imagery. Previous methods that use simple band math analysis have produced promising results but have drawbacks in producing consistent results when low signal to noise ratio (SNR) images are used. Moreover, in seeking to automate the dust storm detection, the presence of clouds in the vicinity of the dust storm creates a challenge in being able to distinguish these two types of image texture. This paper not only addresses the detection of the dust storm in the imagery, it also attempts to find the transport direction and the location of the sources of the dust storm. We propose a spatio-spectral processing approach with two components: visualization and automation. Both approaches are based on digital image processing techniques including directional analysis and filtering. The visualization technique is intended to enhance the image in order to locate the dust sources. The automation technique is proposed to detect the transport direction of the dust storm. These techniques can be used in a system to provide timely warnings of dust storms or hazard assessments for transportation, aviation, environmental safety, and public health.

A low complexity overcomplete directional image pyramid

An undecimated directional filter bank (UDFB) derived from the DFB originally proposed by Bamberger and Smith is proposed. The new UDFB has excellent orientation selectivity, and maintains the low computational complexity from its predecessor. While over complete, the UDFB presents other properties like shift invariance, desirable for some image analysis applications. Using ladder structures the overall directional response is readily controlled by 1D prototypes which are easy to design. We combine the UDFB with over complete pyramidal structures to form directional pyramids with excellent radial and directional selectivity.

Rotation invariant texture classification using Bamberger pyramids

We study the application of the Bamberger directional filter bank to the problem of rotation invariant texture classification. We explore the use of purely directional decompositions and the use of polar-separable Bamberger pyramids. We obtain comparable classification performance to Gabor-based methods using a smaller feature set.

Rotationally-blind texture classification using frame sequential approximation error curves

In this paper we report a new set of rotation invariant features for texture classification. The proposed feature set is based on principles of image approximation using multiresolution (MR) frame decompositions. Features are obtained from sequential approximation error curves (SAECs) obtained from the reconstruction error of texture approximations. These approximations are formed by the sequential addition of frame coefficients in decreasing magnitude order. Feature selection consist of taking points along the SAEC. It is found that SAECs are oblivious to rotation, allowing the generation of rotationally blind feature sets. Hence, the computational complexity of classification systems is reduced by eliminating the need for feature post-processing (e.g., DFT-encoding) to achieve rotation invariance (RI). We test the rotationally-blind feature sets for texture classification using different MR frame decompositions and data sets. We show that the proposed SAEC-based feature set achieve classification rates competitive with other schemes using a smaller feature set.

Achieving idempotence in near-lossless JPEG-LS

The lossless and near-lossless image compression standard, JPEG-LS, while offering state-of-the-art compression performance with low complexity, fails to be idempotent in near-lossless mode (i.e., images degrade upon successive compression/decompressions). This paper identifies the cause and presents two solutions. First it presents a modification to the compressor and decompressor that maintains or improves the error bounds and achieves idempotence. Second it describes a preprocessor that acts upon any image and returns one upon which JPEG-LS does perform idempotently at the expense of doubling the guaranteed error bound on a small subset of pixels (typically below 0.5%).

Affine scaling transformation algorithms for harmonic retrieval in a compressive sampling framework

In this paper we investigate the use of the Affine Scaling Transformation (AST) family of algorithms in solving the sparse signal recovery problem of harmonic retrieval for the DFT-grid frequencies case. We present the problem in the more general Compressive Sampling/Sensing (CS) framework where any set of incomplete, linearly independent measurements can be used to recover or approximate a sparse signal. The compressive sampling problem has been approached mostly as a problem of l1 norm minimization, which can be solved via an associated linear programming problem. More recently, attention has shifted to the random linear projection measurements case. For the harmonic retrieval problem, we focus on linear measurements in the form of: consecutively located time samples, randomly located time samples, and (Gaussian) random linear projections. We use the AST family of algorithms which is applicable to the more general problem of minimization of the lp p-norm-like diversity measure that includes the numerosity (p=0), and the l1 norm (p=1). Of particular interest in this paper is to experimentally find a relationship between the minimum number M of measurements needed for perfect recovery and the number of components K of the sparse signal, which is N samples long. Of further interest is the number of AST iterations required to converge to its solution for various values of the parameter p. In addition, we quantify the reconstruction error to assess the closeness of the AST solution to the original signal. Results show that the AST for p=1 requires 3-5 times more iterations to converge to its solution than AST for p=0. The minimum number of data measurements needed for perfect recovery is approximately the same on the average for all values of p, however, there is an increasing spread as p is reduced from p=1 to p=0. Finally, we briefly contrast the AST results with those obtained using another l1 minimization algorithm solver.