对于噪声大、分辨率低的航拍图像,提出了利用小波分析多尺度特性,在较粗尺度下对图像进行分割的算法,可有效抑制Mean Shift图像分割算法的过分割问题。分析了Mean Shift算法中核函数窗宽变化对图像分割结果的影响,提出了改进的误分率表示方法,利用新的图像分割性能指标,对几幅实际的航拍图像进行了定量分析,表明了新算法的优点并说明了误分率性能评价指标的有效性。
In order to segment the aerial image with large noise and low resolution,a new image segmentation algorithm is proposed,which introduces the characteristic of multiple scales in wavelet to the image segmentation with wide scale and can control the over-segmentation in the Mean Shift image segmentation algorithm efficiently.Moreover,the effect of image segmentation under different kernel wide window is analyzed and a performance evaluation method called segmentation error is promoted.At last,several real aerial images are segmented by new algorithm and evaluated by new method.The result shows the superiority of new algorithm and performance evaluation method.