彩色文件影像分割

Abstract

本篇論文提出了一個可用來分割與辨識彩色文件中之文字和圖形的方法。 首先，利用一平均值分割演算法將彩色影像量化成較少顏色，並分解成一 系列的單色平面。此方法是由中間值切割法改進而來的。藉由此方法，可 適當決定量化後的顏色個數。接著本文提出了一個多頻譜有限制性段長( run-length)演算法來抽取彩色平面中之物件。此一演算法能將各個單色 平面的文字擷取出來，同時將各平面上之非文字區塊組合成完整的圖形。 即使文字落於複雜的背景之上，仍然可以輕易地抽取出來。論文最後提出 了一套新的法則來辨認文字區塊，此方法可彌補統計特徵的不足。實驗結 果證實所提技巧確實可行且具有實用性。 An approach to segmenting and recognizing color document components including texts and graphics is proposed. First, a mean-cut algorithm, which is a revised version of the median- cut method, is used to quantize a color image and decompose the result into a set of planes, each with a single color. By the mean-cut algorithm, a number of reduced colors can be decided appropriately. A multi-spectral constrained run-length algorithm is proposed to extract objects from the color planes. The algorithm can extract texts and merge non-texts of each plane to form graphic components. Even when a textline resides on a complicated background, it still can be taken out easily. Finally, a new method is also developed to recognize textline components. This method can eliminate the shortcoming of using statistical features in textline recognition. Some experimental results are also shown to prove the feasibility and practicability of the proposed approach.