中文摘要：

为了寻求代价更小、效率更高、适应性更强的图像原型表征方法,借鉴成分识别理论的观点,设计出一种更符合人类认知原理、更具有可理解性的物体拟合算法。利用二维高斯混合函数,用高斯成分来拟合物体的边缘图像,使得物体的表征由单一的像素表示转变为利用成分进行表征的方式。为了使得拟合结果更具有健壮性,在算法中还引入了分裂-归约机制来对拟合结果进行修正。实验结果表明,这种拟合手段能够很好地描述物体的特征成分,为图像进行后期的高级语义处理奠定了基础。

英文摘要：

For representing image prototype, minor price, more efficient and more flexible, this paper designs an object fitting algo-rithm which conforms human ′ s recognition mechanism and has much intelligibility based on recognition- by- component theory. The designed algorithm uses mixture of 2- dimensional Gaussian component to fit the object ′ s edge images, and makes object representation from single pixel converted into component. For seek more robust fitting algorithm, a Split- Convergence mechanism is intro-duced to amend the fitting results. The experimental results demonstrated that this fitting algorithm can well describe the object fea-ture component, and laying a good foundation for image high- level semantic processing.