Daniele Casali

Cellular Neural Networks With Virtual Template Expansion for Retinal Vessel Segmentation

A retinal vessel segmentation method based on cellular neural networks (CNNs) is proposed. The CNN design is characterized by a virtual template expansion obtained through a multistep operation. It is based on linear space-invariant 3times3 templates and can be realized using existing chip prototypes like the ACE16K. The proposed design is capable of performing vessel segmentation within a short computation time. It was tested on a publicly available database of color images of the retina, using receiver operating characteristic curves. The simulation results show good performance comparable with that of the best existing methods

Neural associative memory storing gray-coded gray-scale images

We present a neural associative memory storing gray-scale images. The proposed approach is based on a suitable decomposition of the gray-scale image into gray-coded binary images, stored in brain-state-in-a-box-type binary neural networks. Both learning and recall can be implemented by parallel computation, with time saving. The learning algorithm, used to store the binary images, guarantees asymptotic stability of the stored patterns, low computational cost, and control of the weights precision. Some design examples and computer simulations are presented to show the effectiveness of the proposed method.

Associative Memory Design Using Support Vector Machines

The relation existing between support vector machines (SVMs) and recurrent associative memories is investigated. The design of associative memories based on the generalized brain-state-in-a-box (GBSB) neural model is formulated as a set of independent classification tasks which can be efficiently solved by standard software packages for SVM learning. Some properties of the networks designed in this way are evidenced, like the fact that surprisingly they follow a generalized Hebbs law. The performance of the SVM approach is compared to existing methods with nonsymmetric connections, by some design examples

Associative memory design for 256 gray-level images using a multilayer neural network

A design procedure is presented for neural associative memories storing gray-scale images. It is an evolution of a previous work based on the decomposition of the image with 2/sup L/ gray levels into L binary patterns, stored in L uncoupled neural networks. In this letter, an L-layer neural network is proposed with both intralayer and interlayer connections. The connections between different layers introduce interactions among all the neurons, increasing the recall performance with respect to the uncoupled case. In particular, the proposed network can store images with the commonly used number of 256 gray levels instead of 16, as in the previous approach.

SVM based transcription system with short-term memory oriented to polyphonic piano music

Automatic music transcription is a challenging topic in audio signal processing. It consists in transforming the musical content of audio data into a symbolic notation. The system discussed in this paper takes as input the sound of a recorded polyphonic piano music and it produces conventional musical representation as output. For each note event two main characteristics are considered: the attack instant and the pitch. Onset detection is obtained through a time-frequency representation of the audio signal. Note classification is based on constant Q transform (CQT) and support vector machines (SVMs). In particular, in this paper we propose a short-term memory based feature vector for classification. To validate the efficacy of short-term memory, we present a collection of experiments using synthesized MIDI files and piano recordings, and we compare the results with other existing approaches.

Analogic CNN algorithm for estimating position and size of moving objects

An analogic CNN algorithm is proposed for detection of multiple moving objects in high resolution, grey-scale images taken from a fixed camera. The algorithm, based on simple 3 × 3 templates, can be implemented using CNN hardware, providing the real-time operation required in surveillance and traffic control applications. Efficient separation of moving objects from the background is obtained through automatic threshold selection. The performance of the proposed method is shown using real-life indoor and outdoor video sequences. Copyright

A New CNN-based Method for Detection of Symmetry Axis

In this paper a method for symmetry axis detection in binary images is presented. The method exploits the nonlinear dynamic behavior of cellular neural networks (CNNs), in particular the propagation of bipolar waves. The image is represented in polar form, transforming the symmetry with respect to an arbitrarily oriented axis in a vertical symmetry: the position of the vertical axis corresponds to the angle of the original symmetry axis. The parallel CNN architecture is useful to speed up the computation, because of the high computational cost of the task. The proposed algorithm is tested on a real image with good results

Detection of Moving Objects in a Binocular Video Sequence

A moving objects detection algorithm is proposed in order to improve the performance in presence of moving objects appearing close in a 2D image but with different distances from the observer. The method requires two distinct cameras with slight horizontal displacement, giving two video sequences. Frame difference is used to evidence the moving objects from the background in each video sequence. Then a disparity map is computed to measure the distance of each object. Finally, these data are merged by using a clustering algorithm giving the number, size and position of moving objects. Most of the processing can be implemented using cellular neural networks (CNN). We tested this method over several sequences, both indoor and outdoor. Experimental results show a significantly improved discrimination when multiple objects are moving at different distances. Moreover, the use of stereo images can be exploited to reduce noise, improving performances for clustering

A cellular neural network based character recognition system

In this paper we present a character recognition system based on cellular neural networks. As a consequence, we deal with a real-time system. All considered features can be extracted by CNN templates; recognition is merely feature-based, with no need of a learning phase or any kind of memory.

Mental task recognition based on SVM classification

In this paper, a brain/computer interface is proposed. The aim of this work is the recognition of the will of a human being, without the need of detecting the movement of any muscle. Disabled people could take, of course, most important advantages from this kind of sensor system, but it could also be useful in many other situations where arms and legs could not be used or a brain-computer interface is required to give commands. In order to achieve the above results, a prerequisite has been that of developing a system capable of recognizing and classifying four kind of tasks: thinking to move the right hand, thinking to move the left hand, performing a simple mathematical operation, and thinking to a carol. The data set exploited in the training and test phase of the system has been acquired by means of 61 electrodes and it is formed by time series subsequently transformed to the frequency domain, in order to obtain the power spectrum. For every electrode we have 128 frequency channels. The classification algorithm that we used is the Support Vector Machine (SVM).