Cognitive human like vision system
Vision system is the translation on PC of neurophysiology knowledge. In particular, of Hubel and Wiesel work. In 1962, they discovered the importance of angles in higher order animals vision. The system is made of a sequence of programs running on PC having just an images acquisition device. The first group of programs draw out the outline (edge). They focalize on angles survey more than on edges completion. The following programs codify forms. The last programs select information to store. The system learns by itself to recognize every kind of object. Afterwards, it will recognize the object even if it will be partially hidden, deformed, put in a different perspective or in different lighting conditions. Forms are codified to occupy very few memory. Features to recognize thousands of forms can be memorized in a megabyte. This does not means that the forms could mingle. In fact, the number of forms which can be recognized without confusion is huge. The limits are the same of human vision.
To recognise forms, the system needs the same teaching which a person requires. It does not require specific competence or programming ability. Indeed, it does not need any programming. It presents the same limits and the same possibilities of human vision. It detects similarities and differences from form to form by itself. The system can define by itself if differences are essential; if so, it memorise them. If not, the system neglects them. Anyway, if a teacher shows all those particulars, the learning is faster. For closed edges building, it does not matter if a figure is taken well lightened. In fact, the edges are joined together in the memory, starting from imperfect edges. The non essentiality of closed edges brings the possible recognition of a partially hidden object, of focusing it and of isolating every character in a cursive writing. As far as the system developer knows the matter, actual systems commercialised or in phase of development compare groups of memorised features with others drawn out from a presented form. Features are punctual and they vary from group to group of objects. They can be very little matters for finger-prints recognition; bars and hooks for characters recognition; or areas, perimeters, moments of inertia for other objects recognition. Systems not using punctual forms get confused when the number of forms is more than roughly ten. As the neurophysiology of higher order animals and of human beings shows, the system here presented is based on edges angles as essential element (even if not the only one) for forms codification. So, every kind of form is classified with the same features. It means that the software here presented does not require any modification when passing from one application to another. It learns as a human being.
Please contact me.
2. APPLICATION FIELDS
Possible technological applications: alphabetical characters (even handwritten, cursive and not separated one from another) read-out; airplanes, ships, cars recognition; faces recognition; robotic vision; electrocardiograms interpretation; finger-prints recognition.
3. DETAILS ON PARTNERSHIP.
The system developer does not know new computer tools, their power and speed, the most efficient way to build and program them. He does not know the rules driving innovative productions financing. He did not tested market needs, which bring to choose the most convenient application. He looks for a firm complementary to him. It will not have to finance a research (all this had been done yet), but it will have to follow industrial aspects linked to development and production of the software still defined. The firm could be a developer of vision software, or one interested in a specific application. Anyway, it should have some experts in information science engineering, in order to rewrite the software here presented, to make it faster, commercially interesting and to suggest the fitting hardware. For the system developer, three people having the above listed competence could make the most simple applications run (such as the cursive read-out) in less than one year of work.