Abstract: Classification of material is very challenging, time consuming and requires expert material engineers. If a robotic system can be developed which can classify the material automatically then it can be very useful for research laboratories, materialography and industry. This work deals with developing such an automatic classification system which can classify materials based on their microstructure images. The images are used to evaluate a series of feature related parameters and then classify the material based these evaluations. Various methodologies exist to study characteristics of a microstructure image such as texture feature, histogram based feature and statistical analysis of the image. In this work, such algorithms are implemented to develop a database of feature related parameters and then classify material based on nearest neighbor method. A fully automatic system is developed using MATLAB that first places a prepared material sample under the microscope, then captures images from a high resolution camera mounted on the microscope and finally, place the material on the right location based on the classification scheme.
Techniques used for parameter evaluation are:
1. Statistical analysis
2. Histogram based feature analysis
3. Texture based analysis
Co-occurence matrix textual features are defined as
Maximum Probability: Max ij(PI,)
Uniformity of Energy: Σi Σj Pij 2
Inverse difference moment of order k: Σ Σ Pijλ/|i-j|k
Contrast: Σi, j |i-j|k (Pij) λ
Variance : Σ (i-µ i)²Σ j Pij)(Σ j (j µ j)²Σ iPij)
Correlation: Σi Σj (i-µx) (j-µy) Pij / σx σy where µx=Σi iΣj jPij, µy=Σj j Σi i Pij and standard deviations σx =Σi (і-µx)² Σj Pij, σy =Σi (j-µy)² Σi Pi
4. Local feature analysis