基于自适应调节核函数的图像显著区域提取方法

An Adaptive Adjusting Kernel Function-Based Extraction Method for Image Salient Area

目前的图像显著区域提取技术仅针对无噪声图像或者没有分析噪声对提取技术的影响。文章提出一种图像显著区域提取新方法,该方法将自适应调节核函数应用在图像显著区域获取中。根据具体图像像素点与周围小区域的差异性来判断该位置的显著性。差异性是与自适应调节核函数有关的单调下降函数。该算法采用多尺度融合的方法获取整幅图的显著区域,对无噪声图像进行显著区域提取分析取得了较好效果。在图像含噪时与两种现有显著区域获取方法进行比较,实验结果表明该算法同样对噪声具有很强的鲁棒性。

Existing visual area detection technology was often used for noise-free image, and the impact of noise on the detection technology was not analyzed. A new visual salient area detection method for noisy image was proposed in this paper. The adaptive kernel adjusting function in visual area detection was used in our method and the salient property was determined by the dissimilarities between a center patch around that pixel and other patches. The dissimilarity was measured as a decreasing function as adaptive kernel regression. At last, the visual salient area was obtained by multi-scale process. In order to demonstrate the feasibility of our approach, several simulation experiments were done. A good effect was obtained in Visual area detection experiments on noise-free images. Compared with two proposed methods for noisy images, our method owned strong anti-noise characteristics and strong robustness.