乐谱图像倾角快速检测方法

A Fast Algorithm for Music Score Image Skew Detection

计算机光学乐谱识别技术是将传统的纸质型乐谱转化为计算机能够“读懂”的数字音乐,在计算机音乐领域中具有重要的应用价值。乐谱识别系统的输入是乐谱扫描图像,而扫描过程中出现的图像倾斜现象,会给识别过程中的谱线定位和谱段切割带来诸多困难,必须对图像作有效的倾斜校正以保证系统的性能。为此,提出了一种快速的乐谱图像倾角检测方法。该方法首先利用乐谱文档的自身结构特点,对图像进行预处理,滤除乐谱图像中不具备方向性的干扰像素,然后通过多组图像水平投影队列间的交叉相关性计算对倾角进行检测。其特点是在确保检测倾角精度的同时具有非常高的执行效率。实验结果表明这一方法是有效、实用的。

Optical music recognition (OMR) is an application of document image understanding to computer music. As a rule, an OMR system uses a scanned image of music score as input. In order to overcome the obstacles during next processes of staff-line detection and segment, skew detection of score image must be taken before recognition step is executed. In the paper, an approach is put forth to detect and correct skew score image. The algorithm is composed of two parts: the pre-processing of image and skew detection. The former using run-length coding of image removes the redundant pixels that are harmful to detect skew angle. The later employs horizontal projection of adjacent vertical image strips and cross-correlation algorithm to calculate the skew angle. The most advantage of the algorithm is that it takes very short processing time with high accuracy. Such improvement has been evaluated by the experimental results.