基于视觉注意机制的黄绿色苹果图像分割

A segmentation method of yellow and green apple images based on visual attention mechanism

针对近色背景黄、绿色苹果图像难以分割的问题,设计了一种基于视觉注意机制的分割方法。首先基于RGB(Red,Green,Blue)采集图像将R-B、2R-G-B色差分量分别作为黄、绿色苹果图像的颜色特征分量,采用基于频率调谐的显著检测模型(Frequency-tuned salient region detection,FT)算法提取以光线正常区域为主的显著图,然后通过基于标记的分水岭算法处理原图像,再用FT算法提取以高亮区域为主的显著图,将2部分显著图分别进行自适应阈值分割,去除小面积等操作获取其二值图像,最后将2个二值图像合并,由此获得黄色和绿色苹果的果实区域。最后进行本研究方法效果图的主观判断和基于分割误差(Af)、假阳性率(FPR)、假阴性率(FNR) 3个评价指标的定量分析。结果表明该方法能更有效地分割出黄、绿色苹果果实,Af、FPR、FNR分别为8. 1%、10. 56%和10. 18%。

Aiming at the problem that the yellow and green apple images in similar background were difficult to be segmented,a segmentation method based on visual attention mechanism was designed.The components of R,G and B in RGB color space were extracted,and the R-B and 2R-G-B color difference were used as color characteristic components of yellow and green apple images,respectively.The salient graphs of the normal light region oriented fruit based on frequency-tuned salient region detection(FT)were extracted.The watershed algorithm based on the mark was used to deal with the original image,and FT algorithm was used to extract the salient map of the highlighted region oriented fruit.Two parts of salient graphs were segmented using adaptive threshold method,respectively.After the removal of small area and boundary object,two partial binary images were obtained and merged to get the fruit target area of yellow and green apple.Finally,the subjective judgment of experimental results and the quantitative analysis of three evaluation indicators based on segmentation error(Af),false positive rate(FPR)and false negative rate(FNR)were carried cut.This method can more effectively segment yellow and green apple fruits,Af,FPR,FNR were 8.1%,10.56%and 10.18%,respectively.