Features Of Natural Images Research Articles

Augmented reality camera tracking with homographies

To realistically integrate 3D graphics into an unprepared environment, camera position must be estimated by tracking natural image features. We apply our technique to cases where feature positions in adjacent frames of an image sequence are related by a homography, or projective transformation. We describe this transformation's computation and demonstrate several applications. First, we use an augmented notice board to explain how a homography, between two images of a planar scene, completely determines the relative camera positions. Second, we show that the homography can also recover pure camera rotations, and we use this to develop an outdoor AR tracking system. Third, we use the system to measure head rotation and form a simple low-cost virtual reality (VR) tracking solution.

Random Cascades on Wavelet Trees and Their Use in Analyzing and Modeling Natural Images

We develop a new class of non-Gaussian multiscale stochastic processes defined by random cascades on trees of multiresolution coefficients. These cascades reproduce a semiparametric class of random variables known as Gaussian scale mixtures, members of which include many of the best known, heavy-tailed distributions. This class of cascade models is rich enough to accurately capture the remarkably regular and non-Gaussian features of natural images, but also sufficiently structured to permit the development of efficient algorithms. In particular, we develop an efficient technique for estimation, and demonstrate in a denoising application that it preserves natural image structure (e.g., edges). Our framework generates global yet structured image models, thereby providing a unified basis for a variety of applications in signal and image processing, including image denoising, coding, and super-resolution.

Statistical Analysis of Spatial Structure in Natural Images

When a sample of natural images is taken and compared with a set of noise images, the two are obviously distinguishable. Even when the noise images are set to have the same power spectral characteristics as the natural images, there is no doubt which is a normal image and which is a noise image. In the study to be reported, we have examined the nature of the spatial characteristics of natural images that allow them to be so discriminated from noise images. The approach is to process large sets of images belonging to various categories through mechanisms that have some similarities to known operations in biological visual systems. Thus, the images are filtered at various spatial scales and at various orientations; the filter outputs are combined into local energy maps; and features are detected in such processed images. The result of these calculations is the distribution of values of some parameter which describes a particular image characteristics for each set of images. Parameters that could support adequate discrimination between two sets of images, for example natural images and noise images, will have largely non-overlapping distributions. In practice it is found that no simple parameters can distinguish obviously different sets of images, but parameters that encapsulate spatial patterns, especially those related to non-accidental image properties, can do so. It is concluded that filter outputs must be followed by nontrivial spatial operations. Suggestions are made as to what are the most plausible.

Fractal error for detecting man-made features in aerial images

A technique is proposed for aiding photointerpreters in detecting man-made features in aerial reconnaissance images. The technique, which uses a metric called fractal error, is based on the observed propensity of natural image features to fit a fractional Brownian motion model. Man-made features usually do not fit this model well, and consequently the fractal error metric may be used as a discriminant function for detecting man-made scene features.

A Catalog of 1-D Features in Natural Images

This paper explores the local form of actual feature types contained in real images. The local energy feature detector is used to locate points in an image where features are found. An unsupervised neural network is trained to capture the mean luminance values and standard deviations of the luminance values in a small neighborhood of these feature points. This local luminance information is called a feature template. After culling and normalization, we arrive at a catalog of local feature forms for the image. Our experiments indicate that the feature forms are self-similar over different images and across scales. When described by their phase angle, features also show some clustering around a small number of types. The size of the feature catalog is small, and shows promising applications in the area of image compression and reconstruction. Quantization of phase angles around the central angles of clusters yields a catalog of synthetic feature templates that further improves the fidelity of the reconstructed images.

A Catalog of 1-D Features in Natural Images

This paper explores the local form of actual feature types contained in real images. The local energy feature detector is used to locate points in an image where features are found. An unsupervised neural network is trained to capture the mean luminance values and standard deviations of the luminance values in a small neighborhood of these feature points. This local luminance information is called a feature template. After culling and normalization, we arrive at a catalog of local feature forms for the image. Our experiments indicate that the feature forms are self-similar over different images and across scales. When described by their phase angle, features also show some clustering around a small number of types. The size of the feature catalog is small, and shows promising applications in the area of image compression and reconstruction. Quantization of phase angles around the central angles of clusters yields a catalog of synthetic feature templates that further improves the fidelity of the reconstructed images.

The Hermite transform-applications

It is demonstrated how the Hermite transform can be used for image coding and analysis. Hierarchical coding structures based on increasingly specified basic patterns, i.e. general 2-D patterns, general 1-D patterns, and specific 1-D patterns such as edges and corners, are presented. In the image coding application, the relation with existing pyramid coders is described. A new coding scheme, based on local one-dimensional image approximations, is introduced. In the image analysis application, the relation between the Hermite transform and existing line/edge detection schemes is described. It is shown that, by concentrating on more specific patterns, the coding efficiency can be increased since fewer coefficients have to be coded. Meanwhile, sufficient descriptive power can be maintained for approximating the most interesting features in natural images. >