A novel moment-based shape description and recognition technique

The task of recognition of objects from their two dimensional views has been attempted in the past using different techniques. The general requirement has been to represent the original two dimensional iconic image to some compact symbolic description to facilitate matching and storage requirements. Often, the quantitative symbolic description that is needed from an image is more likely to be of the order of tens of real numbers. Another requirement is to recognise as identical two patterns which differ in location, rotation or size. A further requirement is to have description sensitive enough to take care of all the features in an object and robust and flexible enough to disregard the minor differences due to noise and image acquisition system defects. The invariance to rotation and translation is achieved with the use of global techniques such as moments, Fourier descriptors and the cyclic chain codings of polygonal approximation of the objects. The scale invariance is also reported to be achieved with some additional computational cost. Such final description is termed as an n-dimensional feature vector represented as a point in n-dimensional space. Minimum-distance object classification can efficiently be used with such a description