Abstract: Visual saliency detection has very important applications in many aspects such as image segmentation, adaptive compression and object recognition. This paper presents a saliency detection algorithm based on HSV color, texture and spatial position. By this method, the image is divided into small pieces in order to get the local information of the image, and color saliency map is computed in combination with the images’color uniqueness and spatial distribution to compute color saliency map. At the same time, the paper uses Gabor filters at different scales and directions to get the texture feature vector, and then calculates the difference of texture feature vectors to get the texture saliency map. Finally, the combination of the two gets a final saliency map. The experimental results show that this method can get satisfactory results in terms of detection and noise immunity, etc.

Key words: saliency detection, image block, HSV color space, texture feature, target detection

摘要： 视觉显著性检测是很多计算机视觉任务的重要步骤，在图像分割、自适应压缩和识别物体方面都有很重要的应用。提出了一种基于HSV颜色、纹理特征和空间位置关系相结合的显著性检测算法。该方法先将图像分割成小的图像片以获取图像的局部信息，结合图像片颜色的独特性和空间分布的紧凑性计算得到颜色显著图；同时利用Gabor滤波器对图像进行不同尺度和方向地滤波得到纹理特征向量，然后对特征向量计算纹理差异得到纹理显著图；最后将二者结合得到最终显著图。实验结果表明，该方法在检测效果和抗噪能力等方面均可获得较为满意的结果。

关键词: 显著性检测, 图像分块, HSV颜色空间, 纹理特征, 目标检测