Development of a New Algorithm for Segmentation of Flotation Froth Images

"It is well known that froth visual features reflect the operating conditions of the flotation process, so that being able to accurately obtain the froth properties is the most significant criteria to optimize and control this process. Froth segmentation is a useful procedure that can determine the bubble size distribution. Several algorithms have been proposed in this field, but marker-based watershed transform shows the best performance. In spite of this fact, the algorithm suffers from oversegmentation in cases when the flotation froth includes large bubbles along with small ones. In the paper, the marker-based watershed method is modified to prevent oversegmentation of large bubbles. The developed algorithm is validated using some froth images captured at different operating conditions, so the results indicate that the method can segment the mixture of big and small bubbles effectively.IntroductionFroth flotation is a widely used process for separating valuable from gangue minerals. Effective control of flotation circuits has a dramatic effect on overall performance of a mineral processing plant. In most flotation plants, process operators monitor the froth surface visually and make adjustments to process parameters based on their interpretation of the surface appearance and their idea of the appearance of an ideal froth. However, due to natural complexity of the flotation process, achieving optimal control is often not possible for human operators.In recent years, attempts have been made to control the flotation circuits using machine vision systems (Moolman et al., 1995; 1996a,b; Bartolacci et al., 2006; Kaartinen et al., 2006; Vanegas and Holtham, 2008; Aldrich et al., 2010). The froth bubble size distribution is the most significant control variable that reflects the operating conditions in a flotation machine. To date, a number of approaches have been proposed for bubble size measurement from froth images, including segmentation (Sadr-kazemi and Cilliers, 1997; Wang et al., 2003; Mehrshad and Massinaei, 2011), texture spectrum (Nguyen and Thornton, 1995), wavelet (Liu et al., 2005), an improved texture spectrum approach, a method based on binary images (Lin et al., 2008) and, finally, using interfacial morphological information (Yang et al., 2009)."