Article Preview
Top1. Introduction
In the field of image processing, the intensity probability distribution of an input 2-D image is usually represented as a 1-D intensity histogram. As image resolution usually increases due to the significant improvement of digital imaging techniques, the local intensity probability distribution of each patch in the input image needs to be individually estimated. This, in turn, will result in improved effectiveness of the related image processing algorithms, such as data hiding and watermarking (Guo, 2007; Lee et al., 2013; Guo et al., 2009; Wang et al., 2014; Guo et al., 2008; Guo & Liu, 2010; Guo & Liu, 2010), video quality analysis (Huang, 2011), face detection (Guo et al., 2011), image classification (Liu et al., 2011), contrast enhancement (Huang et al., 2013), traffic monitoring (Huang et al., 2013), motion detection (Huang et al., 2014), and so on.
Object detection is a well-known feature generation task that generally is used to estimate the intensity probability distributions in video surveillance systems (Chen & Chen, 2008; Luo et al., 2010). For detection of specific objects or behaviors, such as humans and their actions, the patch histogram is vital to locating the detection region of the input image. In addition to object detection, the video surveillance system also incorporates visual tracking, by which one or more objects are located over time (Medeiros et al., 2010; Nejhum et al., 2010). Based on its generated features, an object can be tracked by measuring the maximum similarity of the local region of the image. Hence, image histograms are applicable as measurement features. As probability or other selective features are generated, target recognition functions use them to classify the specific objects, actions, et cetera (Ning et al., 2009; Ruta et al., 2010).
The amount of noise present in an image is used as the significant index for measuring its quality. Hence, noise reduction algorithms have been proposed for image improvement. For noise estimation or reduction, the patch histogram is often employed (Porikli, 2008; Wan et al., 2010). Image contrast affects visual perception when images are observed. Histogram equalization (HE) is the traditional contrast enhancement method, which maps the curve of the intensity probability density (Liu et al., 2011; Celik et al., 2011). For single image haze removal, the dark channel prior (DCP) method has been the most popular method in recent years (He et al., 2011). The DCP-based defogging algorithms must minimize the gray-level of each pixel to model the illumination and reflectance parameter. As the number of neighbor pixels increases, the process may incorporate the patch histogram construction procedure to reduce computation complexity.