Classifying 3D Real-World Texture Images by Combining Maximum Response 8, 4th Order of Auto Correlation and Colortons

摘要

A novel three-dimensional (3D) texture classification approach is proposed. It combines the use of Maximum Response 8 Filters (MR8) and 4th order of Auto Correlation Filters (Autocor). These features are combined using Between-Class Error (BCE) information that is learned in a Pre-Testing Stage. Variance (MR8) and invariance (Autocor) of rotation present advantages and disadvantages. Our approach is intended to eliminate those drawbacks and combine favorable features. Our approach performs two single classification methods individually and in parallel. It selects output from the method that obtains the fewest BCEs. Additionally, color from the texture images was considered as an effective additional feature for texture classification. The overall performance of our approach was drastically improved by the addition of that feature. The efficiency of our approach was evaluated by employing over 5500 texture images that correspond to 61 real-world surface samples from the Columbia-Utrecht Reflectance and Texture (CUReT) database [2]. A classification percentage of 97.77% for 61 classes was achieved when Autocor [12] and MR8 [13] were combined. Superior performance was achieved by adding new color feature data: Colortons. Using them, the classification rate was 98.94% for 61 classes.