Intensity invariant nonlinear correlation filtering in spatially disjoint noise.

We analyze the performance of a nonlinear correlation called the Locally Adaptive Contrast Invariant Filter in the presence of spatially disjoint noise under the peak-to-sidelobe ratio (PSR) metric. We show that the PSR using the nonlinear correlation improves as the disjoint noise intensity increases, whereas, for common linear filtering, it goes to zero. Experimental results as well as comparisons with a classical matched filter are given.

Improved locally adaptive least-squares detection of differences in images.

We introduce a method for change detection under nonuniform changes of intensity using an improved least-squares method. A locally adaptive normalizing window is correlated with the two images, and a morphological postprocessing is then applied to isolate objects that have been added or removed from the scene. We use a modification of the least-squares solution to get rid of clutter caused by intensity changes that do not satisfy the model assumed for the least-squares solution.

Three-dimensional object detection under arbitrary lighting conditions.

A novel method of 3D object recognition independent of lighting conditions is presented. The recognition model is based on a vector space representation using an orthonormal basis generated by the Lambertian reflectance functions obtained with distant light sources. Changing the lighting conditions corresponds to multiplying the elementary images by a constant factor and because of that, all possible lighting views will be elements that belong to that vector space. The recognition method proposed is based on the calculation of the angle between the vector associated with a certain illuminated 3D object and that subspace. We define the angle in terms of linear correlations to get shift and illumination-invariant detection.

Intensity-invariant nonlinear filtering for detection in camouflage.

We introduce a method based on an orthonormal vector space basis representation to detect camouflaged targets in natural environments. The method is intensity invariant so that camouflaged targets are detected independently of the illumination conditions. The detection technique does not require one to know the exact camouflage pattern, but only the class of patterns (e.g., foliage, netting, woods). We use nonlinear filtering and the calculation of several correlations. The nonlinearity of the filtering process also allows high discrimination against false targets. Several experiments confirm the target detectability where strong camouflage might delude even human viewers.

Nonlinear pattern recognition correlators based on color-encoding single-channel systems.

In color pattern recognition, color channels are normally processed separately and afterward the correlation outputs are combined. This is the definition of multichannel processing. We combine a single-channel method with nonlinear filtering based on nonlinear correlations. These nonlinear correlations yield better discrimination than common matched filtering. The method codes color information as amplitude and phase distributions and is followed by correlations related to binary decompositions. The technique is based on binary decompositions of the red, green, and blue and the hue, saturation, and intensity monochromatic channels of the reference and of the input scene, after which the binary information on the red, green, and blue channels and that of the hue, saturation, and intensity channels are encoded as different angles of a phase distribution. We have applied the method to images degraded by high levels of substitutive noise. Results show that the sliced orthogonal nonlinear generalized correlation detects the target with a high degree of discrimination when other methods fail.

Nonlinear filter for pattern recognition invariant to illumination and to out-of-plane rotations.

Automatic target recognition in uncontrolled conditions is a difficult task because many parametersare involved. This study deals with the recognition of targets under limited out-of-plane rotations while maintaining invariance to ambient light illumination. Contrast invariance is achieved by using the recently developed locally adaptive contrast-invariant filter, a method that yields correlation peaks whose values are invariant under any linear transformation of intensity. To reduce the sensitivity to the orientation of the object we replace the reference in the nonlinear filter by a synthetic discriminant filter. The range used for out-of-plane rotations was 40 degrees with a depression angle of 20 degrees. We present results for unsegmented targets on complex backgrounds with the presence of false targets.

Optical implementation of the weighted sliced orthogonal nonlinear generalized correlation for nonuniform illumination conditions.

Optical pattern recognition under variations of illumination is an important issue. The sliced orthogonal nonlinear generalized (SONG) correlation has been proposed as an optical pattern recognition tool to discriminate with high efficiency between objects. But, at the same time, the SONG correlation is very sensitive to gray-scale image variations. In a previous work, we expanded the definition ofthe SONG correlation to the Weighted SONG (WSONG) correlation to modify the discrimination capability in a controlled way. Here, we propose to use the WSONG when pattern recognition is obtained by means of optical correlation under nonuniform illumination. The calculation of the WSONG correlation requires the summation of many linear correlations between binary images. To implement it optically, we use a time sequential joint transform correlator.

Recognition of unsegmented targets invariant under transformations of intensity.

Images taken in noncooperative environments do not always have targets under the same illumination conditions. There is a need for methods to detect targets independently of the illumination. We propose a technique that yields correlation peaks that are invariant under a linear intensity transformation of object intensity. The new locally adaptive contrast-invariant filter accomplishes this by combining three correlations in a nonlinear way. This method is not only intensity invariant but also has good discrimination and resistance to noise. We present simulation results for various intensity transformations with and without random and correlated noise. When the noise is high enough to threaten errors, the method trades off intensity invariance in order to achieve the optimum signal to noise ratio, and the peak to sidelobe ratio in the presence of clutter is always greater than one. In the presence of random disjoint noise, the signal to noise ratio is independent of the target contrast and of the level of the noise.