In recent years, Vision Systems have found their ways into many applications. This includes fields such as computer graphics, medical, industries such as assembly line inspection and object manipulation. The application of Computer Vision technology to factory automation, Machine Vision, is growing at rapid rate. However in most Machine Vision systems an algorithm is needed to infer 3D information regarding the objects in the field of view.
In this thesis presents an updated Stereo Matching Algorithm of MSOM (Modify Self Organized Maps). This technique based on artificial neural network. Using the learning rule to similarity and differentiation of both sides of the eyes to find the most overlapping position of the images, give the best clarity for human visualization of 3D images, in the machine procedure on stereo vision problems. Feature selection and extraction is an important step for disparity plan. Through studying of several feature detector method found that image gradient filter has direction and can calculate the horizontal, vertical and diagonal direction of the image and provides enough information regarding the critical points that an object can be characterized.
In original MSOM algorithm estimating the disparity and shown the result in gray scale cannot see the distance of the objects. In this method we used a self-adapting dissimilarity measure for extract a disparity plane. Then estimating disparity by update the original MSOM to calculate the disparity and also indicate the depth of the objects in the image. Red color represents nearest object then Green and Blue represent to furthest object.