基于分数阶微分、水平集及分水岭的铅锌浮选图像分割

Flotation Image Segmentation by Combining Fractional Differential,Improved Level Set and Watershed

针对铅锌浮选气泡暗颜色、细节弱、分割难的特征,提出了一种新的图像分割算法。该算法分成3个部分:气泡边界增强,即基于分形学改进分数阶微分算法,根据图像的纹理特性自动确定分数阶微分的非整数阶数,以自适应分数阶微分算法增强气泡边缘;气泡亮点区域提取,即在改进传统的水平集算法基础上,进行气泡亮点区域的精确提取,以克服全局自动阈值算法在提取气泡亮点时存在的缺陷;图像分割,即利用内外标记修正梯度图像,运用分水岭算法对浮选图像进行分割。对不同类型铅锌矿气泡图像进行实验,并通过与多种传统的图像分割算法分析比较。结果表明,新算法不仅提高了浮选气泡图像的分割精度,还有效地减少了传统图像分割算法的过分割问题,本文算法对于浮选气泡具有良好的分割效果。

In order to solve the problem of flotation bubbles, a new algorithm combining the fractional differential, improved level set and Watershed was proposed. The algorithm includes three steps of image enhancement, bubble white spot extraction, and image segmentation. In image enhancement, the fractional differential order was automatically determined on image texture information by using fractal analysis, and the improved adaptive fractional differential algorithm was used to enhance bubble edges. In bubble white spot extraction, the traditional level set algorithm was improved, and it was applied for extracting white spots of bubbles. In image segmentation,the in- ternal and external markers of bubbles were utilized to correct the gradient images, and finally the marked watershed segmentation algo- rithm was used to delineate bubbles. For a number of lead zink ore froth images with different bubble sizes, several traditional image segmentation algorithms were compared with the new algorithm, and the result showed that the proposed algorithm can improve the image segmentation accuracy and is effective in reducing the over-segmentation problem and in segmenting flotation bubbles.