5

I am newbie to the CV field and I am on a relatively simple task, I want to analyse dominoes rock values.

I used blob analysis method, which wasn't really efficient on android for that I used a pure java code which tends to be slower than native C. And background data caused me problems as it could have blobs too.

After too much reading I ran through 'template matching' method which sounds like the best to me, for its low resources need and rotation/scaling tolerance, I can easily save the template of every piece of the dominoes and then match it in the coming image for analysis and counting.

Now all I want is your guidance to libraries/methods that can help me implementing template matching, I prefer pure java but can use native C libs if no choice is found.

P.S: If you find me wrong regarding algorithm I chose, please provide me with your advice.

4

1 回答 1

8

图像识别是计算机视觉中的高级处理，首先您必须进行一些预处理并首先检测对象。

要进行模板匹配，首先必须找到要识别的图像的感兴趣区域。为此，您可以从许多算法中选择一种，例如基于特征的检测和基于外观的检测。找到 ROI（Region of Interest）后就可以进行模板匹配了（我假设你已经理解了模板匹配的概念）

它的摘要你可以在下面的流程图中看到它

模板匹配流程图

和模板匹配的例子

在此处输入图像描述

在您了解整个过程后，您可以使用一些支持 Android 平台的可用库，例如​​高通的OpenCV（开源）或FastCV，它们都使用原生 C，因此您必须使用 android NDK 来编译它。

如果您有任何问题，请随时在评论中问我：）

于 2012-11-05T07:14:19.997 回答