结合熵与局部信息的伪影偏差场修正CV模型

发布时间：2018-05-12 21:11

本文选题：图像分割 + CV模型　；参考：《哈尔滨工程大学学报》2017年05期

【摘要】：针对Chan-Vese(CV)模型对含有伪影、光照不均的图像不能进行有效分割的不足,本文提出了结合熵与局部信息的动态伪影偏差场修正CV模型。模型根据区域同质性特征,利用熵构造区域能量系数,自动调节目标与背景区域在模型中的权重。采用全局与局部结合的方式自适应控制区域演化。将伪影指示函数应用到区域检测项,无需先验灰度信息即可消除异常值,精确地使像素归类。结合Retinex理论对图像进行分解,忽略亮度变化并提取不含照度信息的目标结构图像,避免偏差场对分割的影响。通过与CV模型、LIF模型对比验证了算法的有效性,结果表明,本文提出的算法在目标干扰严重情况下分割性能最优,重叠率可达0.9,误分割率控制在0.06以内。与CV模型、LIF模型相比分割精度与速度性能优势明显。
[Abstract]:Aiming at the deficiency of Chan-Vesegne CV) model which can not effectively segment images with artifacts and uneven illumination, a modified CV model combining entropy and local information is proposed in this paper. According to the characteristics of regional homogeneity, the model uses entropy to construct regional energy coefficient, and automatically adjusts the weight of target and background region in the model. The global and local combination is adopted to control the evolution of the region adaptively. The artifact indication function is applied to the region detection item, and the outliers can be eliminated without prior gray level information, and the pixels can be classified accurately. Based on the Retinex theory, the image is decomposed, the brightness change is ignored and the target structure image without illumination information is extracted to avoid the effect of the deviation field on the segmentation. The validity of the proposed algorithm is verified by comparing it with the CV model / LIF model. The results show that the proposed algorithm has the best segmentation performance in the case of serious target interference, with an overlap rate of 0.9 and an error segmentation rate of less than 0.06. Compared with the CV model and the LIF model, the segmentation accuracy and the speed performance are obvious.
【作者单位】：南京理工大学机械工程学院;
【基金】：国家自然科学基金项目(61105094) 江苏省科研创新计划(CXLX12-0189)
【分类号】：TP391.41

【相似文献】