自然背景下的梅花图像分割算法研究

针对梅花数据的特点，提出一种基于自然背景下的梅花花朵分割算法TC(Texture Color)。该算法综合运用了分形纹理和颜色2种特征，有效分割背景图像中的干扰物，实现梅花图像分割。首先，采用双毯子方法计算图像的局部分形维数图，并对分形维数图采用大津阈值分割去除背景中大部分的干扰物；然后，利用颜色特征对剩余的干扰物进行有效分割。在采用颜色特征进行分割时，改进了色度直方图累加算法，并融合了饱和度特征，取得了很好的分割效果。在分割过程中，算法还采取了形态学操作、去噪和填充等处理技术，得到最终的分割结果。对9种梅花图像（每种20幅，共180幅）进行了分割实验，采用误分率对实验结果进行评价，并分别和2RGB模型分割方法、GrabCut算法进行了实验对比。实验结果表明:TC算法平均误分率控制在3%之内，比2RGB模型分割方法更加有效，并且该算法所耗费的时间比GrabCut算法要少很多而且无需人工交互。因此，本文提出的TC算法针对梅花图像的分割是非常有效的。 

Image segmentation of Prunus mume flower under natural background.

An algorithm T&C(Texture & Color) for automatically segmenting the Prunus mume flower image under natural background was proposed according to the appearance characteristics of P.mume flower.In the algorithm,two features,fractal texture and color were used to segment the interferents in the background.Firstly,the double blanket method was used to compute the local fractal dimension(LFD).Most interferents can be removed using otsu algorithm to segment the LFD image.Then the image was segmented using the relevant information of hue accumulative histogram and saturation histogram.To further improve the segmentation effects,some other operations were used in segmentation process,such as morphological operations,de-noising and filling processing technology.The segmentation experiments and evaluation were made on a database of 180 P.mume images belonging to 9 different types(20 images for each type).The experimental results showed that the average probability of error can be controlled within 3% with our T&C algorithm,which was better than the 2RGB model method.In addition,compared with the GrabCut algorithm,this method has higher performance and need less artificial interaction.So T&C algorithm is effective in segmenting the P.mume image under natural background.