A Shape Feature Extraction Method Based on 3D Convolution Masks

Texture analysis is important in 2D image classification, recognition, segmentation and detection. Although a significant amount of work has been done on 2D image data analysis, techniques for analyzing 3D volume data such as 3D solid textures have not been investigated sufficiently. In this research, we have extended the well-known Laws' texture energy approach to handle 3D solid textures. In our approach, the Laws' texture kernels are convolved together to generate three dimensional masks (3times3times3) while traditional approaches use 2D masks (3times3). The extended 3D Laws' convolution masks make it possible to analyze 3D solid texture databases. Our preliminary experiment shows that the 3D masks are capable of extracting shape features directly from 3D solid textures, although traditional techniques indirectly extract shape features from a sequence of 2D images which are sliced from 3D solid textures. The 3D mask can be used for various 3D solid texture analysis techniques including similarity retrieval, classification, recognition, and segmentation