首页 > 技术文章 > 视频分析(MATLAB)——MV分镜头图像分类

2008nmj 2018-10-08 19:00 原文

引言:一个MV视频是有很多帧图像组合而成的,而一支MV是有多少个分镜头场景组合而成的呢?由MATLAB如何自动实现?

以《Love You Like A Love Song》的MV为例(这里合成帧设置为了12帧/秒,选择合成-->添加到渲染队列-->输出设置为JPEG序列-->点击渲染 一共2640张图片)

导入MATLAB

根据图像相似性进行分类(两两对比,一次对比,如果两帧不相似而且与前面几类都不相似,就新建一类,看总共可以分为几类。每一类有多少帧,哪里帧。)

视频分析-->分类算法-->镜头数

一个镜头里的环境是不变的,人物的妆容的色彩是一样的

Classification Algorithm:

(image classification)

Image Classification of Video Processing

 CNN(FaceDetection) on Video Processing

其实核心还是特征匹配不是吗?相似度评价?

https://baike.baidu.com/item/%E5%9B%BE%E5%83%8F%E5%8C%B9%E9%85%8D/9688914?fr=aladdin

2640帧(Frames)图像,,,,检测出人脸的所有图像,,,,(Before Faceswap)

人脸识别算法(dlib)......

直方图法...计算每一帧的值

Detection of defferent shots in a music video using semantic method

In one music video, there are different shots.But they clipped and combined them together to make a whole music video. If we want to know how many shots they have take. It is hard and tiring for human to detect by human eyes. There are many ways to classify. By Histogram, By Others like histogram, but we propose a new way to classify.by semantic.

CNN is very hot right now.

human detection is also popular.

Method

1. sequence by 1-2640.

2. calculate number

3. ....

train --> ...-->detect-->classification

监督与无监督

推荐阅读