The set of techniques and methods for obtaining classifiers and recognisers has a very solid and well established base. We are witnessing a proliferation of works which update concepts and apply them to real problems. The common factor of these initiatives lies in the fact that society poses new and more complex problems that require new formulas that take into account their specificities. For example, the web is an example of the application domain in which: a) the fruitlessness of potential spaces of representation, b) the vastness of the (representative) sets of examples, and c) the heterogeneity of the different problems that may arise, make any classical solution unsuitable to recognise patterns in this context. If we add applications where textual or hypertextual information is combined with visual information, the level of complexity grows enormously.
The main objectives of this course of action are: the classification methods based on examples (nearest neighbours, support vector machines and RBF neural networks); the selection of attributes and prototypes; multiple classifier systems: bagging, boosting; scalable classifiers; incremental classification and continuous learning systems; heterogeneous clustering; semi-supervised clustering; principal components and independent components in classification; text, hypertext and visual data mining applications; biometric applications; information classification and retrieval in large databases; efficient search and indexing algorithms.

Images are highly redundant because a spatial or temporal portion of the signals can be predicted from its environment. In order to optimise the way in which information is processed, any biological or artificial system dedicated to the analysis of natural images must organise its sensors so as to maximize the independence of its answers. This hypothesis, which links fundamentally the Information Theory, Statistical Pattern Recognition and the computational models of Human Vision, is the central point of this research.
The general objective of this line of work is to deepen the understanding of the statistics of images and natural sequences to explain the functioning of biological vision systems and propose innovative solutions in many applications of image processing and artificial vision where the core of the problem is the selection of an efficient signal representation. The link between the proposed solutions is the inspiration of the algorithms in human vision models assuming that biological solutions have evolved for the optimal treatment of natural images.
The basic and applied areas in which we have been working show this interaction between human visual system models, the statistics of natural images and image processing applications: (i) Image and video compression, (ii) Motion Estimation (iii) Image restoration, (iv) Techniques for image databases and coding (selection of the representation of texture and colour, and distance measures), (v) Characterisation of the human visual system behaviour.

The key in visual information retrieval is based on an adequate representation of the visual content of the images. Low-level visual aspects such as colour, texture, shape, spatial relationships and motion along with other high-level aspects such as the meaning of objects and scenes are used as keys for retrieval of images on the database. The use of visual information in searches arises from the limitations of text-based classic retrieval systems, which are not suitable for modeling the perceptual proximity between images.
The basic theoretical goals of this line are: (i) To propose new low-level descriptors about shape, size and texture based on size distributions on grey-level and colour images, (ii) To develop similarity models inspired on fuzzy logic that incorporate psychological aspects closer to the human judgment of similarity, in collaboration with the RTIS group. The practical objectives are: (i) Applications in ceramics: evaluate the descriptors and the proposed measures on database of ceramic tiles (we chose this application because of its enormous importance in the Valencian Community); (ii) Application to medical images: obtain information to compare a pathology among a specific number of patients with similar characteristics, or to monitor their evolution compared with other patients (particularly in corneas databases).

The fundamental scientific interests of this line are related to the signal acquisition and processing of sensors installed on mobile robots. We are currently working with a prototype built by members of the group, which includes ultrasonic and infrared sensors and cameras. The objective is to determine the location of the robot in its environment, to build maps as large and useful as possible, and to link as directly as possible the sensor signals to the control commands for the actuators, closing the sensorimotor feedback loop at the lowest possible level.
This allows for the integration of the various modules in predominantly active control architectures, though the actual framework of the robot, with a suitable modified core operating system, allows to perform simple tests of randomised control architectures, from the completely hierarchical to the completely reactive ones. These fundamental objectives are translated into other more specific, such as:
(i) The introduction of new types of sensors, or the use of regular sensors in an alternative way. (ii) Research on the processing of visual signals (images) provided by a camera on the robot, using appropriate hardware, specifically programmable logic (FPGAs), which implement useful vision algorithms such as the detection of moving objects (independent of the motion of the robot) or the determination of time-to-impact of the mobile robot with a potential obstacle. (iii) The test of different control architectures. The modification of the Linux operating system core makes it possible to easily test functional, reactive and mixed architectures. (iv) The progressive replacement of classic robot control for neural control, using neural network models (in particular, the CMAC).