基于凸锥模型的图像集分类方法的改进  

An Improved Method for Image-set Classification Based on Convex Cone

作　　者：赵蓉城 ZHAO Rongcheng(Sichuan Business Vocational College,Chengdu,Sichuan 611131)

机构地区：[1]四川商务职业学院,四川成都611131

出　　处：《绵阳师范学院学报》2023年第8期86-96,共11页Journal of Mianyang Teachers＇ College

摘　　要：提出了基于凸锥的图像集分类框架.为了增强框架,引入了两个判别空间.首先提出一个判别空间,使锥之间的差异最大化和类内方差最小化.然后将其扩展到一个加权判别空间,通过引入权重来处理复杂的数据分布.此外,为了减少所提方法的计算成本,开发了一种新的快速实现策略.通过实验验证了所提方法的有效性.A novel convex cone-based image classification framework was proposed.Two discriminant Spaces were introduced to enhance the framework.A discriminant space was proposed to maximize the gap between cones and minimize the variance within the class.Then it was extended to a weighted discriminant space to deal with complex data distribution by introducing weights.Moreover,new fast implementation of the proposed method was constructed to reduce the computational cost.The effectiveness of the proposed method was validated by experiments.

关 键 词：图像识别 凸锥 判别空间 

分 类 号：TP391.14[自动化与计算机技术—计算机应用技术]