A GMM-MRF Based Image Segmentation Approach for Interface Level Estimation

Abstract

Detection of interface level between immiscible liquids has wide applications in various process industries. For instances, fotation cells are used in mining industry to separate components based on froth and liquid layers while primary Separation Vessel (PSV) units are employed in the extraction of bitumen from the oil sands. In all these applications, maintaining the desired interface level between the top froth layer and the liquid layer plays an important role in achieving high recovery of products. As varying throughputs and downstream disturbances tend to change the interface level over time, it is an important indicator of the process behavior. In this paper, we propose an approach based on Gaussian mixture model and Markov Random Field (MRF) based unsupervised image segmentation to achieve the real-time accurate measurement of the interface. The image processing problem is solved as a Maximum a Posteriori (MAP) estimation problem employing the MRF framework and the parameters are estimated using the EM algorithm. The proposed approach is validated using the images captured from a laboratory scale equipment designed to simulate the industrial PSV interface.