Current Research Projects
1 - Populational Biological Monitoring of top of the food chain Pantanal wetland birds
In this project we will build a methodology that will enable the monitoring of colonies of Mycteria americana, which is a bird found in Pantanal, Brazilian wetlands. To that, we will apply image processing techniques to estimate (count) the number of individual from aerial digital images. Provided the investigated bird species lies on the top of the food chain, and is therefore prone to environmental disturbances, we believe that our methodology can be also applied to other animal species of similar characteristics.
2 - Segmentation of large Images by means of Complex Networks
The segmentation of images with community detector algorithms, which are based on Complex Network models, is an appealing alternative when compared to more traditional models. However, it is impaired by very high computation times. This projects seeks to investigates mechanisms which will allow to reduce processing times while preserving the quality and accuracy of segmentation. To that, we introduce the idea of superpixels.
3 - 3D segmentation of the head for assessing bone changes applied to Odontology
Computed tomography plays a key role in the diagnosis and treatment analysis, for example, changes in skull over time for applications in dentistry. This process is conducted by dentists, and consists of the use of tools that manipulate 3D data and generate segmentations of important anatomical structures. However, these tools are not satisfactory, since many targets produce false positives or false negatives. Moreover, it is often necessary to combine the functionality of not just one, but multiple tools to achieve a minimally satisfactory result, which is not a simple task for such professionals. The aim of this PhD project is to define, develop and refine techniques for interactive segmentation of 3D CT scans of the skull to identify important anatomic structures, allowing experts to extract for therapeutic measures. Once the scans to be used in this project have high resolution, segmentation will be performed in three steps: 1) pre-segmentation to reduce the resolution of CT, 2) segmentation of structures main image and 3) interactivity with the user to refine the segmentation. The pre-segmentation techniques will be performed by grouping voxels with similar properties, called Super voxels. Segmentation of structures will be made by means of algorithms for detecting communities in complex networks, and the interactivity with the user is based on pre-segmentation of similar voxels, so as to take advantage of their knowledge to refine the segmentation. It is hoped that this project contributes to the state of the art of 3D segmentation of CT scans of the skull, enhance and develop methods capable of segmenting high-resolution volumes in acceptable computational times.