Tarique da Silveira Cavalcante

Brinell and Vickers Hardness Measurement Using Image Processing and Analysis Techniques

Mechanical hardness testing is fundamental in the evaluation of the mechanical properties of metallic materials due to the fact that the hardness values allow one to determine the wear resistance of the material involved, as well as the approximate values of its ductility and flow tension, among a number of other key characteristics. As a result, the main objective of the present work has been the development and analysis of a computational methodology capable of determining the Brinell and Vickers hardness values from hardness indentation images, which are based on image processing and analysis algorithms. In order to validate the methodology that has been developed, comparisons of the results resulting from the consideration of ten indentation image samples obtained through the conventional manual hardness measurement approach and a computational methodology have been carried out. This analysis allows one to conclude that the semi-automatic measurement of Vickers and Brinell hardnesses by the computational approach is easier, faster, and less dependent on the operator’s subjectivity.

3D Lung Fissure Segmentation in TC images based in Textures

Among all cancers, lung cancer (LC) is the most common of all malignant tumors. In order to obtain a more effective segmentation of pulmonary fissures, independent to other structures present in the CT scan, this paper proposes the segmentation of 3D fissures using texture measures and Artificial Neural Networks (ANN). The results of this study are based on voxels classified as fissure through the proposed method compared to the gold standard set by an expert. The results were analyzed using similarity coefficient rate of 95.6%, the rate sensitivity of 71.1% and specificity of 95.6% rate. Thus, it is possible to identify the job has a gain due to not using segmentation of other pulmonary structures and does not require the use of pulmonary atlas.

Three-dimensional Radial Active Contour Model: A 3-D to 1-D Image Segmentation Technique

In this paper we present a new technique of 3D volume segmentation to be applied in general 3D shapes. This technique, called Three-dimensional Radial Active Contour Model (3D Radial Snakes), generalize the concept of Active Rays and use the idea of rays coming from a central point inside an object of interest to be segmented. This rays lies among a 3D space and are represented by two angles and its length. The idea of this model is to achieve the edge of the object of interest by the analysis of the information only on the ray (1D). We perform tests in synthetic 3D shapes varying the number of rays and computing the mean distance from the original shape and calculating how similar the shapes are. The obtained results from synthetic volumes show that this method can be applied in image segmentation with good accuracy in a fast way.

AUTOIN: Method of Automatic Initialization of Active Contours Applied to Lungs in CT Images

In this work we present a new method of automatic initialization for the active contours method denominated AUTOIN. This method is applied in 72 computed tomography images of the lungs of healthy volunteers and ill patients. AUTOIN is compared to the method of object center localization. The accuracy obtained is calculated using the average and standard deviation of the distance (mm) between the points localized in each method. The results of AUTOIN for healthy volunteers are for the right lung mean of 12.3mm and standard deviation of 17.9mm, and for the left one mean of 11.7mm and 16.7mm of standard deviation. For the ill patients, the results are mean of 15.8mm and standard deviation of 20.4 mm for the right lung and mean of 14.6mm and 21.3mm of standard deviation for the left one. We conclude that AUTOIN enables to perform automatic definition of the initial contour inside the lungs.

New Approach To Automatic Detection Of Strange Objects In Body Scan Images

In this article we present a new approach to automatic detection of strange objects in body scan images specifically in the region of the arms. This methodology is based on a combination of textures, a classifier (K-means or MLP) and a post-processing step. The tests were performed on 23 body scan images of volunteers. The accuracy of this approach is verified by the similarity and sensitivity coefficient with the count of identified and unidentified strange objects. The results indicate that the classifier K-means obtained 92.3% and 78.7% for the similarity and sensitivity coefficient, respectively, while the MLP neural network obtained 100% and 61.9% for the same coefficients. Given these results, it confirms the effectiveness of the methodology and discusses the use of MLP classifier for applications with strict visual inspection stage and the use of K-means classifier in applications where the incidence of false positives hinders the inspection result.