c++ - 如何应用自适应模式生成

Question

其中一篇论文中定义的Moving Object Extraction Method方法如下:

The input of the proposed FLD-based RBF scheme is built in YCbCr color space via three variables regarding perception with which to provide support for a many digital video applications. These variables are luminance (Y), blue-difference chroma (Cb), and red-difference chroma (Cr). Accordingly, the color element of a pixel pt(x; y) uses Y, Cb, and Cr values together to represent the intensity and color of each pixel in each incoming video frame It. In order to provide for the variable bit-rate video stream properties, it is necessary to produce lower-dimensional discriminant patterns. This is achieved through use of the optimal projection vectors though the FLD technique from the continual influx of incoming frames in the discriminant pattern extraction operation. The optimal projection vectors are obtained through a procedure which maximizes the ratio of the between-class scatter and the within-class scatter [32], [33]. The proposed method split each incoming frame into N N blocks, with the kth block xk belonging to the ith class. Let the between-class scatter matrix be determined as follows:

and the within-class scatter matrix determined as follows:

现在有一些我对图像感到困惑的术语:

1. 具有 RGB 值的图像的亮度是多少？
2. 什么是图像类？
3. 什么是图像的散点矩阵？
4. 我如何计算 Ni，u、xk 等是什么？

Answer 1

1. 可以计算相对亮度from linear RGB components :

   Y = 0.2126 R + 0.7152 G + 0.0722 B

   可以找到其他一些从 RGB 空间到 YCbCr 的转换 here .

2. 类可以看作是聚类。通过一些降维和投影的方法，将图像中具有相似模式的 block 聚类为一类，以便在较低的维度上进一步处理。

3. 散点矩阵在 (1) 和 (2) 中定义。它们是反射(reflect)类间和类内相似性的度量。假设这些矩阵用于在提取不同模式的同时对相似模式进行聚类。因此聚类过程是通过调整每个 block 直到SB/SW达到最大值来实现的。

4. Xk为第k block 的亮度，ui为属于第k block 的所有 block 的平均亮度值ith<​​ 类，u 是所有ui 的平均值(所有类的平均亮度)。不确定什么是 Ni(可能是一些与每个类中的 block 数成正比的权重因子)。