區域性二元化圖形對於影像及視訊之分析

Abstract

本篇論文呈現了著名的材質分析運算法：區域性二元化圖形 (local binary patterns，LBP)，其統計圖構成了一個材質分析上重要的特徵。LBP 是由材質區域性的定義所導出，具有不受影像灰階值比例改變而變化的算子；同時它也被發展成擁有不依影像旋轉而改變的特性。且為了適合不同的應用，LBP延伸出各式各樣的形式。 我們不只介紹LBP基礎的理論，更著重於LBP在影像和視訊上的著名應用。材質分析是LBP最傳統的用法，我們評估了多種不同以LBP為基礎，且包含色彩資訊的特徵，來分辨68種不同的材質。臉部偵測及辨識是另一個由於LBP對材質有利的分辨能力而衍生出的應用，我們用同一種特徵來統一臉部定位及辨識。最後，偵測移動物體是LBP最新的應用，經由依序列的可適性LBP統計圖，我們可以建構背景的模型以及分割出有移動的物體。我們實現了這個方法，並給予一些建議。

This thesis presents the local binary patterns, the well-know operator of texture analysis. The LBP histogram constructs an important feature for texture. The operator is defined as a gray-scale invariant texture measure, derived from a general definition of texture in a local neighborhood. It is made invariant against the rotation of the image domain. Various extensions of LBP have been developed to get preferable characteristics. We introduce not only the fundamental theory but also the noticeable applications in image and video analysis with LBP. The texture analysis is a traditional use of LBP. We evaluate the performance of several LBP-based features included color information to classify 68 difference kinds of textures. Face detection and recognition is another utilization of LBP due to the powerful distinguishing ability of texture. We unified both face detection and recognition with the same feature. Finally, the newest application is moving object detection with LBP. Via using a sequence of adaptive LBP histograms, we can model the background and detect objects with movement. This method is implemented and some advice is given.