Introduction to A Nonlinear Feature Extractor for Texture Segmentation Seminar Topic:
Texture segmentation is the process by which images are partitioned into various different and distinct regions or areas. It is extremely useful in many areas such as surface inspection, image processing from satellites, picture archiving, image retrieval using content base etc.
Texture analysis techniques are classified into:
- Statistical
- Structural
- Transform based
- Model Based
In the common texture classification and segmentation process the input image is first transformed and then extracted in various stages. The extracted image is then conditioned and subjected to classification process. After classification we get the segmented image as the output. Various materials such as Gabor filters and wavelets are used the transformation process whereas smoothing filter and a transformation nonlinear functionis used for conditioning.
Finally before segmentation and during the classification stage, feature vectors are given as inputs to the classifier or a neural network which includes various feature images. In this article we are dealing with a network architecture used as an extractor which is nonlinear for the process of texture segmentation. These networks are 2-D and are feed forward. The advantages of these types of networks are it can have different variety of layers based during different texture segmentation. This neural model is represented by a mathematical expression.
The biggest advantage of this segmentation system is that it is independent of filters which are pre-defined. The texture images used for this system are the Brodatz textured images. The feature maps relates to the planes into which extracted layers are arranged, which is in turn distinguished by weight sets.
These weight sets are also known as mask sets will be distributed equally between all units of the map. These weights undergo training by specific set of algorithms which are finally used in the form of convolutional kernels. The pixel approach is always better than a region approach and there is less chance of errors.
Download A Nonlinear Feature Extractor for Texture Segmentation Seminar Report.