弱假设学习在近年来的机器学习领域得到越来越广泛的应用。本文提出一种多分类的基于应力的SBB(Stress Based Boosting)算法。在实时情况下,对输入的英文手语字母进行分类和识别,并转换成相应的汉字,实现了汉字的手势输入。首先对视频流进行预处理,利用双检测方法,消除噪声和背景得到手部区域,使用固定尺度的兴趣点检测器,利用特征提取器归纳出手部特征点,再利用SBB算法的学习规则对含有特征点的新样本进行分析和归类,采用应力反馈对于错判的手势进行纠正。在每个周期,都加入特征学习过程。算法能集中原手势训练集,通过已有分类器的分类结果确定分类策略。本系统具有自主学习、分类快速准确的特点。实验结果证明,本系统在实际条件下具有较高的识别率和鲁棒性,具备很好的应用前景。
Most image interpolation algorithms currently used suffer visually to some extent the effects of blurred edges and jagged artifacts in the image. This letter presents an adaptive feature preserving bidirectional flow process, where an inverse diffusion is performed to enhance edges along the normal directions to the iso-phote lines (edges), while a normal diffusion is done to remove artifacts ('jaggies') along the tangent directions. In order to preserve image features such as edges, angles and textures, the nonlinear diffusion coefficients are locally adjusted according to the first order and the second order directional derivatives of the image. Experimental results on the Lena image demonstrate that our interpolation algorithm substantially improves the subjective quality of the interpolated images over conventional interpolations.
