ROI detection in high speed laryngeal images

The present work describes a new procedure to find the region of interest with a high speed laryngeal video sequence. This approach is based on analyzing the average intensity variations both in the columns and in the rows of the images. The first variations to take into account are those arising in each column. The graphics obtained from these variations will resemble to a Gaussian, where the maximum peak will be the column with the most average intensity variation. In order to determine the cut-off points, the data is fitted to a Gaussian distribution; the mean value will be the maximum intensity variation and the tolerance interval will be used to generate the boundaries for the new image. The procedure described above will be repeated by using the new sequence obtained, but for the average intensity variation in the rows. In this way we obtain the region of interest (ROI). The performance, effectiveness and validation of this approach was proved in 18 high speed digital videos, in which the images have an inappropriate closure of the vocal folds.