影像分割技術在壓縮和醫學影像處理上的應用

Abstract

在過去的幾十年，影像切割已經成為一個從影像處理到影像分析的重要步驟，其主要的目的是要將影像分割成多個種類或區域，其每個種類或區域各自對於一些測量都有相同的性質。而切割的結果對於許多的影像處理加工十分有用，且可以廣泛地應用在各種的研究領域，例如: 醫學影像分析、影像壓縮、物體偵測與匹配、視訊處理等。 隨著醫學影像的大小以及數量逐漸增加，使用電腦自動化地促進影像的處理與分析變得必須。尤其是對於描繪解剖學上的組織結構以及其他感興趣的區域，影像切割演算法在眾多的生物醫學影像應用中扮演著重要的角色，像是量化組織的容積、診斷、解剖學上的組織結構研究、電腦整合手術。 在這篇論文中，我們提出一個運用影像切割的受傷肌肉判斷演算法，此演算法可以直接地從一張超音波肌肉影像中找出健康與不健康的肌肉纖維，並求得一個受傷的分數。根據受傷分數我們可以判斷該肌肉的健康狀況以及估計傳統上藉由染色劑所決定的纖維化程度。模擬結果顯示受傷分數與纖維化有高度的相關性。 除此之外，我們在這篇論文中提出另外一個運用影像切割的生物醫學演算法(稱為細胞計數)，其中我們應用形態學以及優角(大於180度的角)運算來找出細胞壁並分離每個細胞。實驗證明我們的演算法改善了在之前方法所存在的準確度與耗時的問題。 近幾年，形狀自適應影像編碼(Shape adaptive image coding)在許多視覺編碼的應用中已變成主流。形狀自適應編碼的優點在於它可以達到一個更高的壓縮率，這是因為切割出來的影像區域其顏色值有高度的相關性。然而，現存的形狀自適應影像壓縮技術有下列幾個缺點: 高複雜度與低效率的影像編碼。 有鑑於上述壓縮的困境，我們提出一個基於三角形和梯形的二維正交離散餘弦轉換(Two dimensional orthogonal DCT expansion in triangular and trapezoid regions)的壓縮技術。此壓縮技術的概念是根據任何的影像分割區塊可以被視為一個任意形狀的多邊形，而多邊形可以由多個三角形和梯形區域組成。因此，在本篇論文中我們提出了一個三角形和梯形切割演算法。實驗結果顯示由我們的演算法所找出的梯形和三角形可以幾乎匹配影像分割區塊。除此之外，我們的影像壓縮技術相較於JPEG以及其他形狀自適應影像壓縮技術有更好的壓縮效果。

During the past few decades, image segmentation has been an important step from image processing to image analysis. The main purpose is to make a division of an image such that each category or region is homogeneous with respect to some measurements. The segmentation results can be useful for subsequent image processing treatment and widely applied to various researched fields, e.g. medical image analysis, image com-pression, object detection and matching, and video processing etc. With the increasing size and number of medical images, automatically facilitating the image processing and analyzing by computer has become necessary. In particular, as a task of delineating anatomical structures and other regions of interest, image segmen-tation algorithms play a crucial role in numerous biomedical image applications such as the quantification of tissue volumes, diagnosis, study of anatomical structure, and com-puter-integrated surgery. In this thesis, we propose an algorithm of muscle injury determination by image segmentation, which can directly find healthy and unhealthy muscle fibers from an ul-trasound image of muscle, and then derive the injury score. According to the injury score, the healthiness of the muscle can be judged and the degree of fibrosis, which is determined by the conventional method using coloring agent can be also estimated. The simulation results show that the injury score has high correlation with the fibrosis. Besides, another biomedical algorithm (called cell counting) by image segmentation is proposed in this thesis. We apply morphology and the reflex angle operation to find out cell walls and separate each cell. Experiments show that our algorithm improves the existing problems of precision and time-consuming in previous methods. In recent years, shape adaptive image coding has become a mainstream in many visual coding applications. The advantage of shape adaptive coding is that it can achieve a higher compression ratio because a segmented image region has high correlation of color values. However, the existing shape adaptive image compression techniques have the following drawbacks: high complexity and inefficient image coding. Bearing in mind the above obstacle of compression, our image compression scheme based on the two dimensional orthogonal DCT expansion in triangular and trapezoid regions is proposed. The concept of the compression scheme is according to that any image segment can be viewed as an arbitrary polygon and a polygon can be composed by several triangular and trapezoid regions. Thus, the triangular and trapezoid segmentation algorithm is proposed in this thesis. The experimental results show that the trapezoids and triangles derived by our algorithm can nearly match an image segment. Furthermore, our image compression scheme achieves better performance than JPEG and other shape adaptive image compression standards.