查找图像中特定字母的所有出现的好方法是什么?
就像,找到 .png 中所有 A 的位置,无论比例如何,但都正确旋转?
编辑:那些说这是一个比我想象的更困难的问题的人是正确的。我想添加一些细节;我正在寻找一种方法,可以从(黑白扫描手写图像)中找到字母的位置,或者给定字母的图像,找到它的 A 性(WRT 某种字体) 。
我意识到这是一个很大的项目,所以我想知道是否有任何开源库可以做这样的事情,或者一个简单的算法。我尝试过一些角匹配方案,但没有成功。
Like, finding the positions of all the A's within a .png, regardless of scale, but all rotated properly?
Edit: Those who remarked that this is a more difficult problem than I am making it out to be are correct. I wanted to add some detail; I'm looking for a way to either find the position of a letter from a (black-on-white scanned handwritten image), or given an image of just the letter, find the A-ness of it (WRT a certain font).
I realize this is a large project, so I was wondering if there were any open-source libraries that do something like this, or a simple algorithm. I've tried some corner-matching schemes but had no luck.
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(2)
首先,您的问题属于计算机视觉领域(不是图像处理,尽管对于外行来说可能看起来如此)。然后,PNG 图像是自然场景的照片还是扫描的文档会有很大的不同。对于自然图像,您需要检查一般对象检测或特别是文本检测方面的文献。对于扫描文档,您的领域是文档分析(或文档处理)。
无论如何,我认为你的问题可能是可以解决的,尽管解决方案可能比你愿意坚持的更远(在努力和时间上)。
First, your question is in the domain of computer vision (not image processing, though it may seem so for the uninitiated). Then, it makes a big difference whether the png image is a photo of a natural scene, or a scanned document. In the case of natural images you need to check the literature in object detection in general or text detection in particular. For scanned documents your field is document analysis (or document processing).
In any case, I think that your problem is probably solvable, although the solution may be farther (in efforts and time) than you might care to persist.
一种非常好的方法是使用 NGC(非灰度相关)和“A”作为模型。问题是这不是尺度不变的方法。事实上,这个问题可能非常复杂。要解决多个尺度问题,您可以为每个尺度和循环(慢速)生成模型。
Well one very good way to do this is to use a NGC (non-grayscale correlation) with your 'A' as model. The problem is this is not a scale invariant method. Indeed this question could be very complicated. To solve the multiple scales you could generate models for each scale and loop (slow).