Image Processing for Bubble Detection in Microfluidics

Starting from recently years, microfluidics devices have been widely used to build the biomedical devices and micro fuel cells. In the research on microfludics, high speed camera is widely used to record the motion of liquid bubbles in the microchannel, based on which the dynamics of bubbles can be interpreted. However, the mechanical vibration of the system causes the channel to deviate from its original position during the image acquisition process. Figure 1(a) shows a typical image of a silicon microchannel with sidewall water injection, in which the injected water bubble and the skewness of the channel are evident. Hence, it is necessary to perform the skewness and translation correction on the raw images acquired by the high speed camera before the correct information of the bubble trajectory can be obtained from the images. Also, to extract the motion information from a huge number of images, we need to isolate the front bubble object from the background image, which will facilitate the automatic calculation of bubble velocity. The spatial deviation of the image can be decomposed into angular and translational components. In the current project, we will first develop an algorithm to efficiently calculate the angular deviation of the image based on solving a minimization problem using Least Square SVM. Also, the translation correction of image can be accomplished in a straightforward way. After the realignment of the image is finished, the moving bubble can be separated from the background image by using Gaussian mixture model and EM algorithm to classify each pixel in the image as either the foreground or background. The test result shows that the present skewness correction algorithm works very well for the tilt angle less than 20 degree. Also, the background separation algorithm can successfully distinguish the moving object from the background image with still droplets in it for various illumination conditions. The major advantage of the current approach is that no iteration is required and computation is very cheap.