Geometrical Interpretation and Design of Multilayer Perceptrons.

The multilayer perceptron (MLP) neural network is interpreted from the geometrical viewpoint in this work, that is, an MLP partition an input feature space into multiple nonoverlapping subspaces using a set of hyperplanes, where the great majority of samples in a subspace belongs to one object class. Based on this high-level idea, we propose a three-layer feedforward MLP (FF-MLP) architecture for its implementation. In the first layer, the input feature space is split into multiple subspaces by a set of partitioning hyperplanes and rectified linear unit (ReLU) activation, which is implemented by the classical two-class linear discriminant analysis (LDA). In the second layer, each neuron activates one of the subspaces formed by the partitioning hyperplanes with specially designed weights. In the third layer, all subspaces of the same class are connected to an output node that represents the object class. The proposed design determines all MLP parameters in a feedforward one-pass fashion analytically without backpropagation. Experiments are conducted to compare the performance of the traditional backpropagation-based MLP (BP-MLP) and the new FF-MLP. It is observed that the FF-MLP outperforms the BP-MLP in terms of design time, training time, and classification performance in several benchmarking datasets. Our source code is available at https://colab.research.google.com/drive/1Gz0L8A-nT4ijrUchrhEXXsnaacrFdenn?usp = sharing.

Multispectral image denoising with optimized vector bilateral filter.

Vector bilateral filtering has been shown to provide good tradeoff between noise removal and edge degradation when applied to multispectral/hyperspectral image denoising. It has also been demonstrated to provide dynamic range enhancement of bands that have impaired signal to noise ratios (SNRs). Typical vector bilateral filtering described in the literature does not use parameters satisfying optimality criteria. We introduce an approach for selection of the parameters of a vector bilateral filter through an optimization procedure rather than by ad hoc means. The approach is based on posing the filtering problem as one of nonlinear estimation and minimization of the Stein's unbiased risk estimate of this nonlinear estimator. Along the way, we provide a plausibility argument through an analytical example as to why vector bilateral filtering outperforms bandwise 2D bilateral filtering in enhancing SNR. Experimental results show that the optimized vector bilateral filter provides improved denoising performance on multispectral images when compared with several other approaches.

Ontology-based malaria parasite stage and species identification from peripheral blood smear images.

The diagnosis and treatment of malaria infection requires detecting the presence of the malaria parasite in the patient as well as identification of the parasite species. We present an image processing-based approach to detect parasites in microscope images of a blood smear and an ontology-based classification of the stage of the parasite for identifying the species of infection. This approach is patterned after the diagnosis approach adopted by a pathologist for visual examination, and hence, is expected to deliver similar results. We formulate several rules based on the morphology of the basic components of a parasite, namely, chromatin dot(s) and cytoplasm, to identify the parasite stage and species. Numerical results are presented for data taken from various patients. A sensitivity of 88% and a specificity of 95% is reported by evaluation of the scheme on 55 images.

Self-similar random field models in discrete space.

Self-similar random fields are of interest in various areas of image processing since they fit certain types of natural patterns and textures. Current treatments of self-similarity in continuous two-dimensional (2-D) space use a definition that is a direct extension of the one-dimensional definition, which requires invariance of the statistics of a random process to time scaling. Current discrete-space 2-D approaches do not consider scaling, but, instead, are based on ad hoc formulations, such as digitizing continuous random fields. In this paper, we show that the current statistical self-similarity definition in continuous space is restrictive and provide an alternative, more general definition. We also provide a formalism for discrete-space statistical self-similarity that relies on a new scaling operator for discrete images. Within the new framework, it is possible to synthesize a wider class of discrete-space self-similar random fields and texture images.