基于IMFO-Otsu的果实深度图像多阈值分割

Depth Image Segmentation of Multilevel Threshold based on Improved Moth Flame Optimization

为了解决传统果实图像进行阈值分割易受颜色、光照等因素影响的问题,提出一种基于改进飞蛾火焰算法(Improved Moth flame Optimization,IMFO)的多阈值分割算法(IMFO-Otsu)。算法在构建深度直方图后,根据多阈值Otsu准则获取最佳分割阈值。为了提高获取最佳阈值的计算效率,对多阈值Otsu准则进行剪枝处理,并使用提出的改进飞蛾火焰算法对算法进行加速。为验证IMFO-Otsu算法的效果,使用该算法对采集得到的果实图像进行多阈值分割,结果表明提出的算法具有良好的性能。由于提出的算法没有用到彩色图像的颜色信息且简单有效,能在夜间环境等复杂情况对果实识别与定位提供支持。

In order to solve the problem that traditional fruit images for thresholding are easily affected by color and illumination,a multi-threshold segmentation algorithm(IMFO-Otsu)based on Improved Moth flame Optimization(IMFO)is proposed.The algorithm obtains the best combination of thresholds according to the multi-threshold Otsu criterion by constructing a depth histogram.To improve the computational efficiency of obtaining the best threshold combination,the multi-threshold Otsu criterion is pruned and the algorithm is accelerated using the proposed improved moth flame algorithm.The results of the multi-threshold segmentation of the acquired fruit images using the IMFO-Otsu algorithm show the good performance of the proposed algorithm.Since the proposed algorithm does not use color information of color images and is simple and effective,it provides a basis for fruit recognition and localization in complex situations such as nighttime environments.