Designing spectral sensitivity curves for use with Artificial Color
Pattern Recognition
Image and Vision Computing
Fuzzy Aggregation with Artificial Color filters
Information Sciences: an International Journal
| Hi-index | 0.00 |
In this dissertation two novel approaches in color image processing and pattern recognition are described: Artificial Color and Margin Setting. Artificial Color is a subset of the problem called Artificial Perception. The idea is to use what is known about Natural Color vision to design machine vision systems. Artificial Color is a discriminant generated by computers and attributed to the object using data obtained through measurements employing multiple overlapping spectral sensitivity curves. It has numerous advantages that are explored in detail here. The design, training, display, and use of Artificial Color systems are discussed. Margin Setting is a new statistical pattern recognition approach. Margin is the room for variation. It seeks very simple separation surfaces and finds a subset of the training set that is classified by that surface with a preselected margin. These two recently introduced techniques are explored jointly in this dissertation. Using Artificial Color means to construct image plane filters. Then the logic operations can be performed on those filters before applying them to scenes. This color segmentation approach can not only make crisp decisions but also has sufficient flexibility for processing of uncertainty so that it can retain as much color information as possible. Several possible different fuzzy T-norms are applied to Artificial Color filter to illustrate the richness they introduce. Using Margin Setting to train the filters allows us to be very conservative in what is definitely assigned to a class while allowing a useful gradation of membership. This dissertation also shows Margin Setting has one primary degree of freedom—the margin. As Artificial Color filters are binary, they too allow various degrees of freedom such as the ability to use median filters of various sizes. Jointly, these parameters allow an Artificial Color filter to be optimized for a specific task. It is proved here that Artificial Color filtering can provide an orthogonal discriminant to the spatial pattern discriminant in iris recognition and searching. It is also shown how to combine results from the two discriminants in such a way as to improve performance of the combined system over either part—something that has been troubling until now.