Optimal echocardiographic image segmentation using statistical texture analysis

This paper deals with an automatic segmentation of echocardiographic images based on a textural approach. We propose an original method to determine optimal texture attributes in each textured region of the image. During a training step, the Karhunen-Loeve Transform (KLT) is computed on the three different pre-defined areas on the image. From these transforms, three sets of optimal F.I.R. filters are deduced. These filters sets applied on an image to process give selective responses according to the texture and allows to automatically classify each pixel into one of the three predefined classes.