视频场景识别请教！！！

不会帮顶 通过『我邮2.0』发布

友情顶贴，估计得要海洋场景的一些模板吧，匹配之类的

快播的人说了，没有这种技术[ema9] 【 在 skyee 的大作中提到: 】 有木有人会识别出视频里面的特定场景啊，比如说在一段视频...

还是CNN啊

本质上还是图像识别理解

最常见的就是bag-of-words feature。。。然后加svm分类。。。

摸摸头，太TM有道理了 【 在 leadersire 的大作中提到: 】 : 快播的人说了，没有这种技术 : 有木有人会识别出视频里面的特定场景啊，比如说在一段视频...

I am not working on the scenery recognition, but there are some ideas and researches I know. I only can just give you a big picture, hope these tips could help you: First of all, I suggest you to read the the literature review on scenery recognition and find what you peer did on this topic; Then, the first method comes to my mind is segmentation. This is a supervised learning. There are a lot of segmentation algorithms that work well today. Pick one of them and apply it to the video stream. Label the region and extract the feature (color, sift, HBR...) and put it into whatever classifier (svm/random forest/bayesian network). If you consider the timeline of the video, you can fit the model into the Markov model, which maybe more difficult. The second method is using the attributes. Davi Parikh has done a lot of work on this topic. Notice that this work may be not applicable to the real-scenery recognition. But definitely it can be used in searching. The last one I think about is the easiest and the most powerful. You can fit the deep neural network on ur problem. Good luck.

楼上屌爆了！ 发自「贵邮」