A novel polar-based human face recognition computational model

Motivated by a recently proposed biologically inspired face recognition approach, we investigated the relation between human behavior and a computational model based on Fourier-Bessel (FB) spatial patterns. We measured human recognition performance of FB filtered face images using an 8-alternative forced-choice method. Test stimuli were generated by converting the images from the spatial to the FB domain, filtering the resulting coefficients with a band-pass filter, and finally taking the inverse FB transformation of the filtered coefficients. The performance of the computational models was tested using a simulation of the psychophysical experiment. In the FB model, face images were first filtered by simulated V1- type neurons and later analyzed globally for their content of FB components. In general, there was a higher human contrast sensitivity to radially than to angularly filtered images, but both functions peaked at the 11.3-16 frequency interval. The FB-based model presented similar behavior with regard to peak position and relative sensitivity, but had a wider frequency band width and a narrower response range. The response pattern of two alternative models, based on local FB analysis and on raw luminance, strongly diverged from the human behavior patterns. These results suggest that human performance can be constrained by the type of information conveyed by polar patterns, and consequently that humans might use FB-like spatial patterns in face processing.

Neural cell classification by wavelets and multiscale curvature.

A new approach to automatic classification of retinal ganglion cells using multiscale techniques including the continuous wavelet transform, curvature, and standard pattern recognition techniques is described. Each neural cell is represented by its outer contour, and the wavelet transform is calculated from the complex signal defined by the aforementioned contour, leading to the so-called W-representation (Antoine et al. 1996). The normalized multiscale wavelet energy (NMWE) is used to define a set of shape measures associated with the number of details of the shape for a broad range of spatial scales. Next, the more discriminating NMWE coefficients are chosen through a feature ordering technique and fed to statistical classifiers. In addition, the normalized multiscale bending energy (NMBE) is discussed as a means of neural shape description for classification purposes based on the multiscale curvature, i.e. the curvegram, of the neural contour. It is shown that both shape descriptors are suitable for shape classification, presenting similar classification performance. In fact, NMBE has a slightly better recognition rate than NMWE in our experiments. On the other hand, NMWE is less computationally expensive than NMBE, presenting also the potentially useful property of allowing the use of more suitable different analyzing wavelets, depending on the problem under consideration. Therefore, both measures are related and provide a good framework for the design of neural cell description and classification. The methods described in this work have been successfully applied to the classification of two classes of cat retinal ganglion cells, namely alpha and beta (henceforth referred as alpha-cells and beta-cells, respectively), and three statistical classifiers were considered: minimum-distance, k-nearest neighbours and maximum likelihood. The mean recognition rates are near 90%, which is superior to the other shape measures considered. It is argued here that the proposed technique can be adopted as a new general methodology for multiscale shape analysis and recognition, being applicable also to other problems in biological shape characterization in neuroscience and general biomedical image analysis. In the context of analysis of shape complexity, the multiscale energies are coherent with subjective judgements by humans.