结合过渡区的多元线性回归昆虫图像分割方法

Insect Image Segmentation Algorithm Based on Multiple Linear Regression and Transition Region

针对多元线性回归模型对含阴影的昆虫图像边界分割不准确的问题,提出一种结合过渡区的多元线性回归优化算法。算法首先对多元线性回归模型进行范数优化。即根据图像的RGB三色板信息建立多元线性回归基本模型,再利用余弦范数对模型进行优化。优化后算法对图像的分割效果有所改进,但仍保留了图像阴影部分,因而引入过渡区算法对边界和阴影进行分割,实现图像边界的二次分割优化。与单独应用多元线性回归算法相比,新算法提高了昆虫图像的分割精度,具有较强的鲁棒性。

Aiming at the drawbacks that the multiple linear regression model is not ideal to the boundary segmentation for insect image with shadow,a new insect image segmentation algorithm based on multiple linear regression and transition region was proposed in this paper.The method first optimizes the multiple linear regression model with norms,i.e.builds the multiple linear regression model based on RGB three color boards of the image,then acquires the regression model optimized by the cosine norm to segment the insect images.The optimized algorithm can improve the image segmentation effects,but it also retains the image shadow.So the transition region algorithm was introduced to segment the boundary from the shadow,and achieve the boundary optimization.Compared with the algorithm which only uses multiple linear regression,the proposed method improves the accuracy of insect image segmentation and has strong robust-ness.