Detection of small bowel tumors in endoscopic capsule images by modeling non-gaussianity of texture descriptors
This paper presents an approach to the automatic detection of small bowel tumors by processing endoscopic capsule images. The most significant texture information is selected by using wavelet processing and captured in the image domain from an appropriate synthesized image. Co-occurrence matrices ar...
| Autor principal: | |
|---|---|
| Outros Autores: | , , |
| Formato: | article |
| Idioma: | eng |
| Publicado em: |
2010
|
| Assuntos: | |
| Texto completo: | http://hdl.handle.net/1822/17771 |
| País: | Portugal |
| Oai: | oai:repositorium.sdum.uminho.pt:1822/17771 |
| Resumo: | This paper presents an approach to the automatic detection of small bowel tumors by processing endoscopic capsule images. The most significant texture information is selected by using wavelet processing and captured in the image domain from an appropriate synthesized image. Co-occurrence matrices are used to derive texture descriptors by modeling second order statistics of color image levels. These descriptors are then modeled by using third and fourth order moments in order to cope with distributions that tend to become non-Gaussian especially in some pathological cases. The proposed approach is supported by a classifier based on radial basis functions procedure for the characterization of the image regions along the video frames. The whole methodology has been applied on real data and shows that higher order moments can be effective in modeling capsule endoscopic images regarding tumor detection. |
|---|