使用 MATLAB 进行物体识别
您对使用 MATLAB 实现 2D 对象识别有什么想法吗？物体的哪些特征可以输入神经网络？
Have you any idea about implementing 2D object recognition with MATLAB? Which characteristics of objects can feed a neural network?
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
对象识别通常是所有图像处理任务的主要部分之一。从简单的案例开始，例如 指纹识别 和 最佳字符识别到运动跟踪等。
当然，考虑到给定的情况，有许多不同的方法 问题。只能基于颜色（颜色基础对象识别），假设有一个红球在绿色区域中，因此只有通过检测绿色，您才能识别球就像这里的跟踪球！另一种简单的方法是形态运算符。此外，可以使用直方图并从像素分布中找到图片的所有边缘 就像这里。此方法也用于查找书面文本的基线。
更高级的方法基于机器学习方法。 神经网络是最著名的，基本上你通过一堆例子来训练你的模型，找到合适的权重/神经元的值，最后要求模型判断新的例子（测试）。当然，向网络提交图像并不是明智之举。除了计算方面之外，还存在过度拟合问题。因此，提取图片中的共同模式是另一个挑战。比如说，所有字符“A”都遵循一些模式，可以是曲线、角度、强度、FT值，并将其与“L”等区分开来。这部分也称为降维，因为您将所有图片像素映射/组合为多个数据点。 PCA（主成分分析） 并检查 PCA 和 SVD 在 matlab 中。这些方法仅在某些最高变异基础上解释数据的变异。
如今更热门的机器学习的另一个观点是统计方法，将对象视为信号，将其他一切视为噪声。 高斯过程 和 内核方法放在这一边。只需在“Google 中的对象跟踪”旁边搜索这些术语即可获得数千篇学术论文！
总之，这是一个非常普遍的问题；根据问题的不同，解决方案可以更改。我建议您查看 https://stats.stackexchange.com/ 因为您的问题适合这些人的专业知识。
Object Recognition is generally one of the main part of all image processing task. From Simple cases, like fingerprint recognition and Optimal Character Recognition to movement tracking and etc.
Of course there are many different approaches, considering given problem. It can be only based on color (color base object recogniton), say there is a red ball in a green field so only by detecting green color you can identify the ball like here for tracking ball!. The other simple approach is Morphological Operator. Furthermore, one can uses histogram and from distribution of pixels, find all edges of the picture like here. This method is also used to find the baseline of the written text as well.
More advanced methods are based on Machine Learning Approach. Neural network is most known which basically you train your model by bunch of example, find proper weights/values for neurons and finally asking the model to judge about the new example (test). Of course submitting an image to the network is not really wise; Apart from the computational aspects, there is an over fitting issue. So extracting common pattern among pictures is another challenge. Say, some pattern that all characters "A" are following, could be the curve, angles, intensity, FT values and distinguish it from "L" and so on. This part is also called as Dimension Reduction, since you are mapping/composing all picture pixels into several data point. PCA (principle component analysis) and also check the PCA and SVD in matlab. These methods explain variation of data only in some most high variant basis.
The other point of view of Machine Learning which these days is more hot Is in Statistical approach, by looking at the object as the signal and everything else as noise. Gaussian Process and Kernel Methods are placed in this side. Just search these terms beside "object tracking in Google to get thousands academic paper!.
In sum, its really general question; Depends on the problem, solution can be changed. I suggest you to look through https://stats.stackexchange.com/ since your question can fit to those guys expertise.