Rene Jaime-Rivas

System fusion in passive sensing using a modified hopfield network

Abstract We address a new approach to the problem of improving the quality of remote-sensing images obtained with several passive systems, in which case we propose to exploit the idea of neural-network-based imaging system fusion. The fusion problem is stated and treated as an aggregate inverse problem of restoration of the original image from the degraded data provided by several image-formation systems. The non-parametric maximum entropy regularization methodology is applied to solve the restoration problem with the control of balance between the gained spatial resolution and noise suppression in the resulting image. The restoration and fusion are performed by minimizing the energy function of the multistate Hopfield-type neural network, which integrates the model parameters of all sensor systems incorporating a priori and measurement information. Simulation examples are presented to illustrate the good overall performance of the fused restoration achieved with the proposed neural network algorithm.

Holistic cursive word recognition based on perceptual features

This work presents a holistic system for the off-line recognition of cursive words based on the extraction of perceptual features by means of convolution with templates of line segments. The method works directly on gray-level images to avoid losing information due to binarization. Features are represented spatially in order to preserve topological information which is well suited for recognition through convolutional neural networks. Moreover, our feature extraction method does not need to detect baselines to find ascenders and descenders. The system has been tested on small and medium vocabulary sizes demonstrating the feasibility of the proposed method.

Cursive word recognition based on interactive activation and early visual processing models.

We present an off-line cursive word recognition system based completely on neural networks: reading models and models of early visual processing. The first stage (normalization) preprocesses the input image in order to reduce letter position uncertainty; the second stage (feature extraction) is based on the feedforward model of orientation selectivity; the third stage (letter pre-recognition) is based on a convolutional neural network, and the last stage (word recognition) is based on the interactive activation model.

Modality control of an active camera for an object recognition task

In this paper, we show an active object recognition system. This system uses a mutual information framework in order to choose the optimal parameters of an active camera for recognizing an unknown object. In a learning step, our system builds a database of all objects by means of a controlled acquisition process over a set of actions. These actions are taken from the set of different feasible configurations for our active sensor. Actions include pan, tilt and zoom values for an active camera. For every action, we compute the conditional probability density of observing some features of interest in the objects to recognize. Using a sequential decision making process, our system determines an optimal action that increases discrimination between objects in our database. This procedure iterates until a decision about the class of the unknown object can be done. We use the color patch mean over a region of interest in our image as the discrimination feature. We have used a set 8 different soda bottles as our test objects and we have obtained a recognition rate of about 99%. The system needs to iterate about 4 times (that is, to perform 4 actions) before being capable of making a decision.

Rotation invariant image recognition using hough transform and support vector machines

The Hough transform is an information preservation transformation which converts rotations of the image into translations. For this reason, it constitutes an ideal candidate for adding rotation invariance to SVM based image recognition systems. Experiments performed on face recognition are presented. The problem of translation invariance is also dealt with.

Fuzzy models for system identification

A computerized environment for the automatic synthesis of a fuzzy model from numerical evidence is introduced. Such a fuzzy model (a controller or decisional one) is a binary-input single-output Mamdani type model. The main task is to adequate the model output to a system output sampled for some input-output relational values called training data. Thus, the model is a fuzzy approximator for the transfer function with description abilities. Fuzzy approaches are used for both the structure identification and optimization. Synthesized models are evaluated in practical cases.

Remotely sensed image fusion with dynamic neural networks

Presents the dynamic Hopfield-type multistate maximum entropy neural network (MENN) for image restoration with data-controlled system fusion. The optimal fusion was accomplished by processing the data provided by several imaging systems incorporating measurements, system calibration and image model information. Applying the developed new aggregation method we performed an optimal adjustment of the parameters of the MENN algorithm by simultaneously controlling the data acquisition balance and resolution-to-noise balance in the fused restored image. Due to this applied system aggregation method the developed MENN exhibited substantially improved resolution performance if compared with those with existing neural-network-based and traditional regularized inversion techniques, which do not accomplish the system fusion tasks.

Texture discrimination through fractal geometry

A method is presented for modeling and identifying textured images. The method is based on the use of iterated function systems, which are utilized for representing the image, and also for getting characteristic measures for different textures. A set of this kind of functions is chosen such that each one of them keeps the relationship between a small region and a larger one, and the regions are selected so that the smaller is inside the larger, for taking the measures about self-similarity properties in that part of the image. These measures are then translated to a feature map, like in the self-organizing map methods, to analyze them.

3-D polyhedral object recognition using fuzzy indicators

Fuzzy logic has been used during the past years for determination and analysis of visual properties and spatial relations in object regions, a fundamental key for the object recognition problem. Here we present a method which quantifies the number of sides of each region of an object of a light intensity image, a measurement that can be used as a helpful aid to identify objects. We add this new property, the fuzzy number of sides, to other fuzzy measurements commonly used in object identification, such as area, diameter, roundness, and elongatedness, and we obtain a useful improvement in reliability for object recognition.

Linear and nonlinear optical characterization of PMMA clusters with Ni nanoparticles dispersed

We analyze linear and no linear optical properties of PMMA clusters in thin film s with Ni nanoparticles dispersed with different concentrations. Saturable absorber and negative nonlinear refraction index behavior evidences were found using z-scan technique. We also show that these properties have not dependence of the type of matrix but they have on concentration.