利用特征场进化的图像分割方法

Feature Field-based Evolution Method for Image Segmentation

针对图像分割中最优阈值的选择问题,在数据场机制的基础上提出了一种高维图像分割方法。将局部灰度特征与Tamura纹理度量相结合,如粗糙度、对比度、方向度等,尽可能提取足够的图像信息;将每个包含多特征的像素视作具有一定质量的质点,在图像特征空间上建立特征场;假设最优阈值为潜在的进化方向,通过因动态数据场中相互作用所导致的质点之间自适应吸引和排斥实现特征场的协同进化,并利用少数服从多数的投票法则确定最终分割结果。实验结果表明,所提出的方法在不明显增加时间耗费的情况下,取得了较好的分割效果,具有合理性和有效性。

In order to correctly select the optimal threshold for image segmentation,a novel method of high-dimensional thresholding based on data field mechanism was proposed.Combing the local grayscale and Tamura's texture model,including coarseness,contrast and directionality,enough image information is extracted as far as possible.Taking each pixel with multi-features as a data particle with certain mass,a feature field is generated in the image feature space.Assuming the optimal threshold is latent optimal evolution direction,and feature field-based coevolution is achieved by adaptive attraction and repulsion among particles because of the interactions in dynamic data field,and the final result is determined by majority rule.The experimental results suggest that the new method obtains better performance without obviously increasing the time cost,and it is efficient and effective.