Color co-occurrence matrix based froth image texture extraction for
mineral flotation
Weihua Gui, Jinping Liu, Chunhua Yang, Ning Chen
⇑
, Xi Liao
School of Information Science and Engineering, Central South University, Changsha 410083, China
article info
Article history:
Received 9 September 2012
Accepted 28 March 2013
Available online 2 May 2013
Keywords:
Froth flotation
Froth image
Color co-occurrence matrix
Gray level co-occurrence matrix
Texture complexity
abstract
It is well accepted that the surface texture appearance of the flotation froth involves crucial information
about its separation process, which can be used as an effective criterion for the qualitative assessment of
the flotation performance. To obtain the distinctive characteristic of the froth surface appearance under
various production conditions, a texture feature extraction method based on color co-occurrence matrix
(CCM) is presented compared to the commonly used gray level co-occurrence matrix (GLCM). First, the
HIS (Hue, Saturation and Intensity) color space is employed to exhibit and quantify the froth image,
which yields a more intuitive description of the color properties in comparison with the RGB (Red, Green
and Blue) color space. Then, the CCM is computed and the corresponding feature statistics of the froth
surface texture are extracted based on the proposed matrix. Next, a new feature parameter is defined
and extracted to describe the froth texture complexity based on the aforementioned texture feature sta-
tistics. After adequate offline froth images have been obtained from a bauxite flotation plant located in
China under various production statuses with the corresponding concentrate grade in the froth assayed
manually, the qualitative relationship between the texture complexity and the corresponding concen-
trate grade is investigated. Consequently, the optimal texture complexity range to achieve satisfactory
production index is obtained for the further research of the optimal control of the flotation process.
Experimental results have verified the effectiveness of the method and demonstrated its superiority over
the previous texture feature extraction methods based on GLCM.
Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction
Froth flotation is one of the most broadly used mineral separa-
tion methods in the mineral processing industries. It aims to sepa-
rate valuable minerals from useless materials or other materials
based on the difference in wettability of different minerals (Aldrich
et al., 2010). During the flotation process, hydrophobic particles
tend to adhere to the air bubbles in the solution, which rise up
to the top surface of the slurry and then can be separated as froth
(Hu, 1991). Due to the inherent chaotic property with the frequent
and random disturbances to flotation circuits, the flotation process
control is far from satisfactory since it remains a poorly understood
process and lacks for generally useful mathematical modeling
(Kaartinen et al., 2006). Even worse, the key metallurgic parame-
ters reflecting the flotation performance, such as concentrate grade
and tailing grade, cannot be detected in real time. Though some
on-stream or in-stream analysis sensors such as X-ray fluorescence
(XRF) Analyzers can be used to measure the mineral contents in
the slurries to a certain extend, there still exists the problem of
detection delay with low detection frequencies and high cost of
maintenance of these instruments (Haavisto and Kaartinen,
2009). Recently, the froth surface appearance has been known as
an effective indicator of the flotation production performance
(Bonifazi et al., 2002). Furthermore, it has long been acknowledged
that better understanding of the change of the froth surface
appearance in the flotation process is key to understanding the
overall behavior of froth flotation systems (Shean and Cilliers,
2011). Hence, the practical operation variables such as chemical re-
agent, air flow and pulp level are mainly adjusted through the
observation of froth surface by experienced operators (Yang
et al., 2009a; Xu et al., 2012).
However, many problems exist in the naked-eye observation
based flotation process operation and control: (1) Human obser-
vation is qualitative and lacking in quantitative measurement of
the froth surface. The evaluation criteria of the froth surface
appearance vary significantly. Even for the same froth flotation
production status observed at the same time, different operators
might report quite different perceptional results. (2) Operators
cannot monitor the whole flotation circuit effectively because
the whole flotation circuit is a long and continuous process
including dozens of flotation cells. Furthermore, operators are
inevitably tired of the long-term observation and cannot accu-
rately identify the froth state with the instantaneous observation
0892-6875/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.mineng.2013.03.024
⇑
Corresponding author. Tel.: +86 13875915950.
E-mail address: ningchen@csu.edu.cn (N. Chen).
Minerals Engineering 46–47 (2013) 60–67
Contents lists available at SciVerse ScienceDirect
Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng