Abstract: Concerning the poor performance and high computation complexity of a single fractal feature applied in detecting Region Of Interest (ROI) of man-made objects, an algorithm for detection of ROI based on fractal and wavelet was proposed. Firstly, the original image was decomposed to sub-images. Secondly, new fractal feature of low frequency sub-image was computed utilizing fractal intercept feature and fractal fitting error, thus ROI of low frequency sub-image could be gained. Finally, ROI of original image was obtained using the relationship between original image coordinates and sub-image coordinates. The experimental results show that the algorithm proposed has very good effect in detecting ROI of image and reducing computational complexity.

Key words: image recognition, region of interest, target detection, fractal features, wavelet transform

摘要： 针对分形应用于人造目标作为感兴趣区域的检测中,单一分形特征检测效果不好及计算量大的问题,提出了一种基于分形和小波相结合的图像感兴趣区域提取算法。首先对图像进行小波分解,然后综合利用低频子图像的分形截距特征和拟合误差特征,得到一个新的有效的分形特征参数,从而检测获得低频子图像的感兴趣区域,最后根据原始图像与子图像的坐标对应关系,得到原始图像的感兴趣区域。实验结果表明,该算法能够有效地检测图像感兴趣区域,并且检测速度快。

关键词: 图像识别, 感兴趣区域, 目标检测, 分形特征, 小波变换