识别位图中的徽标 - android
我想检测用户相机视图中是否有徽标图像 - 例如 StackOverflow 的徽标是否打印在纸上并通过相机查看。或者对于不太复杂的方法，假设我想在位图中检测它。我想知道是否有一个特定的徽标 - 可以将其硬编码到应用程序中。
是否可以使用 OpenCV 等开源库，或者是否需要大量的工时来实现？
I'd like to detect if a logo image in user's camera view - for example if the logo of StackOverflow was printed on a paper and viewed through camera. Or for less complicated approach let's say that I'd like to detect it in a bitmap. I'd like to recognize if one specific logo - it's possible to hardcode it to the app.
Is it even possible with using of open-source libraries like OpenCV or would it require lots and lots of man-hours to implement?
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）
这是不变矩匹配的一个明显例子（查看关于 Hu Moments 的维基百科）。我正在开发免费的开源库，它使用它在 android 上进行文本识别：
http://sourceforge.net/projects /javaocr/
有android demo，演示完整的往返——采样、训练和识别。这是展示它的市场应用程序：
https://market.android.com/details?id=de.pribluda.android.ocrcall&feature=search_result#?t=W251bGwsMSwxLDEsImRlLnByaWJsdWRhLmFuZHJvaWQub2NyY2FsbCJd
（通过邮件列表支持库）
我会说，只对一个人实施认可徽标将花费我大约 5-6 小时。
添加：OpenCV是一个用C编写的大库，将其集成到android中就是主要的PITA
This is a clear case for invariant moments matching (look into wikipedia about Hu Moments). And I'm developing free opensource library which uses it for text tecognition on android:
http://sourceforge.net/projects/javaocr/
There are android demos, demonsrtating complete roundtrip - sampling, training and recognition. Here is market application showcasing it:
https://market.android.com/details?id=de.pribluda.android.ocrcall&feature=search_result#?t=W251bGwsMSwxLDEsImRlLnByaWJsdWRhLmFuZHJvaWQub2NyY2FsbCJd
( Support for the library through mailing lists )
I would say, that implementing recognition for just one logo will take me anout 5-6 hours.
Addition: OpenCV is a big library writen in C, and just integrate it into android is major PITA
这对于 opencv 来说是可能的，但可能仍然需要大量的工时，具体取决于您对批次的定义。
更严重的是：从模板匹配开始。如果这对您来说还不够，也许 SIFT 功能。另外，搜索 这个 网站，有一些关于像这样的东西被发现。
This is possible with opencv but might still require lots and lots of man-hours depending on your definition of lots and lots.
On a more serious note: start at template matching. If that is not enough for you perhaps SIFT features are. Also, search this site, there are some pretty good discussions about stuff like this to be found.