基于多分辨率阈值的非均匀光照微视觉图像实时分割技术研究

摘要

In order to meet the needs of high-precision and real-time target information extraction from micro-vision images in the micro-nano and biological fields, a kind of real-time segmentation method of micro-vision images with uneven illumination based on multi-resolution threshold was proposed. Because the traditional methods are difficult to estimate the threshold accurately in micro-vision image with uneven illumination and consume too much time, firstly, the energy function of gray intensity distribution was established according to the sparseness property of the gradient probability distribution, and the optimum gray intensity distribution was estimated by applying the iterative re-weighted least squares (IRLS) technique for the purpose of uneven illumination compensation in micro-vision image. Secondly, on the basis of the traditional two-dimension Otsu method, multi-resolution analysis of wavelet transform was employed to reduce effectively the computation of threshold estimation process and improve real-time segmentation of the micro-vision images. Experimental results showed that the proposed method possessed the fast and precise segmentation performance, which was also free of the uneven illumination effect and able to coordinate effectively the contradiction between the accuracy and speed of segmentation.%为了满足微纳和生物医学领域对微视觉图像目标信息高精度实时提取的需求,提出一种基于多分辨率阈值的非均匀光照微视觉图像实时分割技术.针对传统方法所存在的在非均匀光照微视觉图像分割过程中难以精确估计阈值和实时性差等问题,首先根据微视觉图像的梯度概率密度稀疏分布特性,建立灰度强度分布估计目标函数,并利用迭代加权最小二乘法实现灰度强度分布优化估计,达到非均匀光照补偿的目的;然后,在传统二维Otsu方法的基础上,利用小波变换的多分辨率分析能力有效减少阈值估计过程中的计算量,提高微视觉图像分割的实时性.实验研究表明,该方法能有效协调分割精度和速度之间的矛盾关系,具有不受光照影响的特点,能快速取得精确的图像分割效果.