智慧型的彩色影像處理

Abstract

近年來隨著科技產品的數位化，越來越多的東西需要影像處理的技巧，不管是在相機、手機、投影機、印表機、液晶電視、LCD…等，都需要將影像或影片美好的表現出來或紀錄下來。另外，自動影像辨識也是影像處理領域中的ㄧ項重要課題，像是安全保密系統，需要辨識一些生物特徵，就需要先將ㄧ張影像做分析，抽取特徵向量及參數，再配合一些演算法加以比對，而獲得想要的結果。 本篇論文主要分為三部分，在第一部分中，用一種稱之為經驗模分解(EMD)的演算法來分析信號及影像，可以將信號拆解成幾個函數(IMF)，進而想利用這些函數的局部特徵來幫助影像辨識的應用。 本論文的第二部份是在探討圖像色彩傳遞的演算法，在環境的限制下，常常某些風景或場所的光源或顏色，所造成的景象，並不讓人們滿意，因此有了顏色傳遞的演算法，利用兩張圖像，ㄧ張原始的彩色影像及ㄧ張想要的目標彩色影像，而可以將原始影像的顏色氣氛轉變為目標影像的，除了可以讓人們將圖片轉成自己想要的顏色之外，亦可以拿來當作顏色矯正之用，或是使用名畫當目標影像，就可以使自然的圖像變成有藝術風格的感覺。 最後一部分，主要是在研究如何將ㄧ張影像，利用現有的資訊，來將它加強，使其更加清晰，或是看起來更加愉人。在許多情況下，相機無法像人眼一樣，在較暗的環境下，紀錄我們所看到的影像時，通常會較暗且有點糢糊，而這裡我們利用Retinex及 LCRT演算法，應用到電腦中的圖案，使它看起來較清晰且保留住原有的局部對比，這樣便可以彌補環境的缺憾，一樣能擁有美好的照片。除此之外，我們在照相時，數位相機都會先經過類比轉數位及壓縮的過程，過程中會有雜訊的產生，造成圖像失去一些資訊，所以要利用bilateral filtering 來降低雜訊，且保留住原有的邊界資訊，邊界輪廓資訊在影像中是很重要的ㄧ環，若使其模糊將無法辨認物體。

With the digitization of the scientific and technological products in recent years, more and more things need the image processing skill, no matter in camera , cell-phone , projector , printer , liquid crystal TV , LCD, etc., they all need to show good-looking images or movies. In addition, the automatic image recognition is also an important subject in image processing, such as a security system that needs to recognize some biometric characteristics. However, we should analyze image and extract some characteristic features before recognition. This thesis is composed of three parts. In first part, we discuss so-called Empirical Mode Decomposition (EMD) algorithm and apply it to analyze signals and images. EMD can decompose the signal into several Intrinsic Mode Functions (IMFs). In second part, we discuss color transfer algorithms. Under the restriction of the environment, pictures we take are not satisfied by people due to the light source or color of scenes at that place. Color transfer algorithms not only transfer color atmosphere from image to image but also can correct scene color. The last part discusses the issue that how to enhance images. In many situations, the camera is unable to imitate human eyes. For example, the camera can’t work well under darker environment. Here we utilize Retinex and LCRT algorithm to make pictures look clearer and keep the local contrast. In addition, when we take pictures by digital camera, the process of analog data to digital data would produce noise. Bilateral filtering can reduce the noise and keep the information of edges.