Multiple Resolution Images Research Articles

Combining boundary and region features inside the combinatorial pyramid for topology-preserving perceptual image segmentation

Combinatorial pyramids represent the image as a stack of successively reduced combinatorial maps, which encode the whole image at different levels of abstraction. Within this framework, this paper proposes to conduct the perceptual organization of the image content in two consecutive stages. The first stage builds the lower set of levels of the hierarchy according to simple face (regions) features (colour and size). On the top of this hierarchy, the second stage will mainly employ boundary features, encoded in the darts of the combinatorial maps, to obtain a second set of levels of abstraction. The Berkeley data set BSDS300 is used to quantitatively compare the performance of the proposal to a number of perceptual grouping approaches, showing that it yields better or similar results than most of these algorithms while offering two interesting features: computation at multiple image resolutions and preservation of the image topology.

Natural scene recognition using weighted histograms of gradient orientation descriptor

The automatic recognition of the contents of a scene is an important issue in the computer vision field. Though considerable progress has been made, the complexity of scenes remains an important challenge to computer vision research. Most of the previous scene recognition models are based on the so-called “bag of visual words” method, which uses some clustering method to quantize the numerous local region descriptors into a codebook. The size of the codebook and the selection of initial clustering center have great influence on the performance. Furthermore, the big size of the codebook has high computational cost and memory consumption. To overcome these drawbacks, we present an unsupervised natural scene recognition approach that is not based on the “bag of visual words” method. This approach works by creating multiple resolution images and partitioning them into sub-regions at different scales. The descriptors of all sub-regions in the same resolution image are directly concatenated for support vector machine (SVM) classifiers. To represent images more effectively, we present a new visual descriptor: weighted histograms of gradient orientation (WHGO). We evaluate our approach on three data sets: the 8 scene categories of Oliva et al., the 13 scene categories of Fei-Fei et al. and the 15 scene categories of Lazebnik et al. Experiments show that the WHGO descriptor outperforms the classical scale invariant feature transform (SIFT) descriptor in natural scene recognition, and our approach achieves good performances with respect to the state of the art methods.

Contour Detection and Hierarchical Image Segmentation

This paper investigates two fundamental problems in computer vision: contour detection and image segmentation. We present state-of-the-art algorithms for both of these tasks. Our contour detector combines multiple local cues into a globalization framework based on spectral clustering. Our segmentation algorithm consists of generic machinery for transforming the output of any contour detector into a hierarchical region tree. In this manner, we reduce the problem of image segmentation to that of contour detection. Extensive experimental evaluation demonstrates that both our contour detection and segmentation methods significantly outperform competing algorithms. The automatically generated hierarchical segmentations can be interactively refined by user-specified annotations. Computation at multiple image resolutions provides a means of coupling our system to recognition applications.

Quality Determination of Steel Surfaces Based on Best Feature Selection

We propose a quality classification methodology based on optimal textural feature selection. This method employs the wavelet packet transform to decompose the original image into multiple-resolution images. Wavelet texture analysis is applied to extract quality-related features from subimages. Optimal textural feature selection is employed to select the discriminative texture in accordance with class information. The previously used best basis approach is incapable of optimal texture classification when combined with wavelet texture analysis. The proposed texture classification method unifies the best basis approach with wavelet texture analysis. Further, we improve the previous best basis to obtain an optimal basis using a simple rule to select discriminative signatures. The proposed methodology is applied and validated for classifying the surface quality of rolled steel sheets. Experimental results show that features extracted using the proposed method are more discriminative than those obtained using the best basis in terms of classification performance and Fisher's index.

A Pixel-Parallel Self-Similitude Processing for Multiple-Resolution Edge-Filtering Analog Image Sensors

A pixel-parallel self-similitude computation architecture has been developed for multiple-resolution directional edge filtering based on focal-plane image processing. The self-similitude organization has enabled pixel-by-pixel multiple-resolution image filtering with minimal complexity in interconnects. As a result, it has become possible to accomplish any (1/2)n-resolution directional edge filtering in (n+2) steps. An analog edge-filtering chip implemented using a switched floating-gate MOS (neuMOS) technology that is capable of performing four-directional edge filtering at full, half, and quarter resolutions was designed and fabricated in a 0.35-mum 3-metal CMOS technology. The concept has been experimentally verified using the fabricated chip, and the four-directional edge filtering at full, half, and quarter resolutions was demonstrated by measurement.

A Procedure to Extrapolate Vegetation Cover Estimates Over Large Arid and Semi‐arid Regions Using Multiple Spatial Resolution Imagery

Typically, land degradation on arid and semi‐arid military training and testing lands in the desert Southwestern United States is associated with a decrease in total vegetative cover. Long‐term field surveys have been established by the military to measure and monitor total vegetative cover over time. However, field methods are not cost effective due to the large area of training lands, nearly 3 million acres in the California deserts alone, that must be sampled. Measurements recorded at individual transects must still be spatially extrapolated to produce a complete census of the installation. Remote sensing has been used to spatially extrapolate such field measurements over larger areas, but in the past, imagery has lacked sufficient spatial resolution to accurately estimate total vegetative cover. In this study, nested, high resolution imagery was collected at different spatial resolutions for study sites at the Marine Corps Air Ground Combat Center in the south‐central Mojave Desert of southern California. Vegetation cover estimates derived from these multiple resolution images were examined as a possible surrogate for sampling total vegetative cover in the field. Results indicate that a high correlation exits between field measurements of cover and estimates of cover derived from imagery. However, similar to field measurements, if the intent is to estimate total vegetative cover across large geographic areas, high resolution imagery is also costly in terms of collection, processing, and interpretation because a single image generally covers a small geographic area. Therefore, a protocol to scale detailed observations of total cover from high resolution imagery to lower resolution imagery that covers a larger geographic area was developed and tested. The results of this study indicate that a relatively high correlation exists between vegetation cover estimates derived from high resolution imagery and image brightness derived from a nested, lower resolution image. This protocol now allows military land managers to sample site‐specific areas with high resolution imagery and extrapolate absolute estimates of vegetation cover across training and testing lands with more cost effective, lower resolution imagery that provides complete coverage of an installation.

Efficient multi-Gaussian expansion of galaxies

We describe a simple, efficient, robust and fully automatic algorithm for the determination of a Multi-Gaussian Expansion (MGE) fit to galaxy images, to be used as a parametrization for the galaxy stellar surface brightness. In most cases the least-squares solution found by this method essentially corresponds to the minimax, constant relative error, MGE approximation of the galaxy surface brightness, with the chosen number of Gaussians. The algorithm is well suited to be used with multiple resolution images (e.g., Hubble Space Telescope [HST] and ground-based). It works orders of magnitude faster and is more accurate than currently available methods. An alternative, more computing intensive, fully linear algorithm, that is guaranteed to converge to the smallest Chi^2 solution, is also discussed. Examples of MGE fits are presented for objects with HST or ground-based photometry, including galaxies with significant isophote twist.

Simultaneous acquisition of multiple resolution images for dynamic contrast enhanced imaging of the breast.

An imaging technique is described that allows the reconstruction of a series of images at high temporal rates, while simultaneously providing images at high spatial resolution. The method allows one to arbitrarily choose from among several combinations of temporal/spatial resolutions during postprocessing. This flexibility is accomplished by strategically interleaving multiple undersampled projection reconstruction datasets (or subapertures), in which each set can be used to reconstruct a high temporal resolution image. Images with increasingly higher spatial resolutions can subsequently be formed by combining two or more subaperture datasets. The technique is demonstrated in vivo to assess the kinetics of contrast enhancement and to visualize the architectural features of suspicious breast lesions.

High-Definition Image Mosaicing Using Multiple Resolution Images and Its Interactive Display System

High-Definition Image Mosaicing Using Multiple Resolution Images and Its Interactive Display System

Multiple resolution image restoration

A new algorithm for solving the deconvolution problem is proposed. This algorithm uses the wavelet transform to induce a multiresolution approach to deconvolve a blurred signal/image. The low resolution part of a signal/ image is restored first and then high resolution information is added successively into the estimation process. Two 'different ways to incorporate the image space positivity constraint, namely loosely and strictly, are discussed. In contrast to most restoration algorithms, the positivity constraint is applied directly in the transformed domain. The performance of the algorithm in the presence of noise is also investigated. © lEE, 1997.

Optical Haar wavelet transforms of binary images

Two architectures in space and frequency domains are given to optically implement wavelet transforms (WT) in real time and in 2-D parallel, which in principle can circumvent the 4-D display requirement for 2-D WT. Specifically, we have experimentally performed the 2-D Haar WT of binary images directly in the space domain by means of a shadow-casting system using 2-D lenslet arrays and micropolarizers. Shadowing is natural for scale changes, and polarization encoding is necessary to realize the bipolar nature of Haar wavelets. Haar wavelets have two elementary types in 2-D, a corner mother wavelet and an edge mother wavelet. Both are useful for the real-time feature extraction for multiple-resolution image processing and pattern recognition. Moreover, a holographic processor that implements the 2-D Haar WT through filtering operations in the frequency domain is numerically simulated. The feasibility of both architectures is demonstrated and compared by computer simulations and experiments.