Combining Gradient and Albedo Data for Rotation Invariant Classification of 3D Surface Texture
ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
International Journal of Computer Vision - Special Issue on Texture Analysis and Synthesis
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
Malignant melanoma (MM) is a life-threatening dermatological disease, the successful treatment of which relies heavily on early recovery and diagnosis. It is proposed that early MMs tend to have larger 3D topological magnitude and direction disruptions than benign lesions. With a view to capturing these 3D indications, a 6 light photometric stereo device has been developed to recover reliable 3D skin texture in the form of surface gradients. In order to characterise the 3D skin texture, two separate numerical methods have been proposed which quantify surface direction and magnitude disruption. A true 2D centroid which is the projection of the real 3D centroid on the image plane has been used to determine the reference angles and the surface direction disruption using minimum error estimation. This is a more accurate method than the conventionally defined centroid which is susceptible to intensity distortions. Preliminary studies on 30 lesions (of which 5 are MMs) have resulted in 80% specificity. This is better than those achieved through analysis of 2D skin line patterns 56% specificity providing the same sensitivity. This demonstrates the proposed 3D texture analysis can provide potentially very useful MM indicators in addition to existing 2D features.