Use of laser speckle and entropy computation to segment images of diffuse objects with longitudinal motion

A system using laser speckle effect is proposed to segment images reflecting vibration movements of di use targets. Longitudinal movements are difficult to identify when simple imaging systems are used. The proposed system produces a two dimensional segmentation of the target and it is sensitive to longitudinal movements. The speckle effect, produced when coherent light is reflected and interferes when hitting rough surfaces, can be used in order to accomplish this purpose. A pattern with high and low intensity spots is observed depending on the illuminated scene. In our optical system, two silicone membranes are illuminated using a beam expanded laser source and their patterns are recorded using a video camera. One of the membranes experiences a longitudinal controlled movement while the remaining scene is still. Speckle data is processed using a temporal gradient and a regional entropy computation. This method produces a binary individual pixel classification. Four sets of parameters have been tested for the entropy computation and the area under the receiver operating characteristic (ROC) curve was used to select the best one. The selected set-up achieved a ROC value of 0.9879. A data set with 12 different membrane velocities was used to define the threshold that maximizes the classifier accuracy. This threshold was applied to a validation data-set composed by 4 sinusoidal movements with distinct velocities. The accuracy of this technique has achieved values between 92% and 97%. The results show that the target was accurately identified with the optical non-contact apparatus and the developed algorithm.