modeling for the human vision system
interneuronic connection constructing
non-uniform neural network
mosaic of cones
M. V. Kostyukov, Yu. N. Pronkin
The human vision system investigation can be considered as vital importance of basic science and is actual for physiology, cognitive psychology, cybernetics, the theory of signal processing and other branches. It turns out that the transformation of visual signals taking place in the human retina is far more complex than simple linear filtering. While studying the human retina it is widely used the computer modeling. Neural network modeling based on physiological data cellular level obtains a particular interest of scientists.
The urgent need for a software system to work with human retina models examining intercellular interactions can be explained by the dissatisfaction with scattered papers and unflagging interest of scientists in the field.
Various chemical and physiological processes occur in the retina and the three main of them are the object of interest. They are the process of light transformation into electrical activity occurring in photoreceptors, the processes of forward and backward (lateral) transmission of electrical signals in the neural network.
Neural network models of the retina have common features. That fact enables to work with them using generalized para-digm. Such a paradigm can be a consideration for the neural network model like a graph of connections consisting of model neurons and model connections. The model neuron is a functional element of the model neural network. It changes its state due to mathematical law. Interaction between model neurons occurs through model connections and also is described by the mathematical equations.
The user is allowed to work with the graph of connections throw the signal flow diagram that determines only the interaction between neuron types. The benefit of doing this is that quite precise behavior of both individual and collective responses of cells may be modeled at a level that is intermediate between detailed cellular-level physiology and macroscopic function. Therefore, the graph of connections is built automatically.
The modeling process for early human vision based on intercellular interactions requires solving several important problems including the problems for specifying stimuli of different types, optical projection calculation, geometric packing formation of photoreceptors and neurons with non-uniformities, the formation of non-uniform intercellular connections, saving the response of the output layer as images, fitting the model parameters. The problem of constructing heterogeneous mosaic of cone cells is highlighted and the original algorithm for solving it is introduced.
The final two sections provide a logical organization of software system with a brief description of its components. Its options are considered. Every logic module is implemented as a class that performs interface functions. Third party libraries were actively (aggressively) involved in to solve some problems, namely they are Boost, CVODE, CGAL and OpenCV. We should emphasize the presence of the module for fitting parameters in automatic mode which can be used if the adjustment of certain parameters of the model is required.
There are possibilities of the software system to work with some discrete models of the retina, the ability to use complex models of neurons such as Hodgkin-Huxley model and the facilities to replace existing modules with user ones.