由於科技的發達,對於攝影機所拍攝的影像,需要有快速良好的處裡方法。本文著重於對攝影機所取得的樂譜影像加以處裡並且播放。主要實現兩個部分。第一、由於一般的辨識軟體,無法針對被扭曲的影像,進行良好的修正。所以,本文特地針對被扭曲的影像,提出新的方法,以便還原成可辨識的影像。首先將影像切割成小方塊,接著使用霍夫變換法計算每塊切割圖的斜率。當取得斜率後,利用三角函數取得方塊偏移量。最後將相鄰且相連的方塊,進行偏移量修正。驗證結果良好,可有效提高圖形辨識率。第二、一般的樂譜辨識軟體,對於音樂符號的辨識法,有採用分割法,也有採用形態學法,但都無法處裡切割時所造成的偏差。為了有效辨識樂符,且為了讓辨識率上升,因此決定使用類神經網路,來提高樂譜辨識的效果,經過多張樂譜影像驗證,證明成功率平均可達96.4%,效果良好。最後,本文將兩個系統整合成一個完整的流程,經由實際讀取樂譜驗證後,證明本文所提出的影像扭曲修正與類神經樂譜辨識方法,確切可行。
As the technology developed for the camera capturing images, people need faster and smarter method for processing. This paper focuses on the image processing of music scores captured from the video camera. There are two main goals. One, because the existing recognition software could not provide a good correction for the distorted image, the paper proposed an improved method. First of all, we cut the images into small pieces. Then, we used the Hough Transform to calculate the slope of each piece. When we obtained slope, we used tangent function to obtain the offset value. Finally, based on the slopes and the connections between the pieces, the correction makes images become better. According to the good results, this process could effectively improve the pattern recognition rate. Two, to effectively identify the music symbols and increase the recognition rate, we decide to use neural networks to enhance the music score recognition. After many experiments were done, proving success rate averaged 96.4%, the results indicate the significant improvements of the music score recognition. Finally, this article will integrate the two approaches into a complete processing method and to verify by the actual reading music score image. Then, we prove the method that distorted image correction and neural network for recognition in this paper is work correctly.