The similarity analysis is a major issue in computer vision. This concept is denoted by a scalar which designates a distance measure giving the resemblance of two objects. Specifically, this distance is used in many areas such as image compression, image matching, biometrics, shape recognition, objects recognition, manufacturing industry, data analysis, etc. Several studies have shown that the choice of similarity measures depends on the type of data. This paper presents an evaluation of some similarity measures in the literature and a proposed similarity function taking into account image feature. The features concerned are textures and key-points. The data used in this study came from multispectral imaging by using visible and thermal infrared images.
Today, most advanced mobile phone cameras integrate multi-image technologies such as high dynamic range (HDR) imaging. The objective of HDR imaging is to overcome some of the limitations imposed by the sensor physics, which limit the performance of small camera sensors used in mobile phones compared to larger sensors used in digital single-lens reflex (DSLR) cameras. In this context, it becomes more and more important to establish new image quality measurement protocols and test scenes that can differentiate the image quality performance of these devices. In this work, we describe image quality measurements for HDR scenes covering local contrast preservation, texture preservation, color consistency, and noise stability. By monitoring these four attributes in both the bright and dark parts of the image, over different dynamic ranges, we benchmarked four leading smartphone cameras using different technologies and contrasted the results with subjective evaluations.
Smartphone cameras have progressed a lot during recent years and even caught up with entry-level DSLR cameras in many standard situations. One domain where the difference remained obvious was portrait photography. Now smartphone manufacturers equip their flagship models with special modes where they computationally simulate shallow depth of field. We propose a method to quantitatively evaluate the quality of such computational bokeh in a reproducible way, focusing on both the quality of the bokeh (depth of field, shape), as well as on artifacts brought by the challenge to accurately differentiate the face of a subject from the background, especially on complex transitions such as curly hairs. Depth of field simulation is a complex topic and standard metrics for out-of-focus blur do not currently exist. The proposed method is based on perceptual, systematic analysis of pictures shot in our lab. We show that the depth of field of the best mobile devices is as shallow as that of DSLRs, but also reveal processing artifacts that are inexistent on DSLRs. Our primary goal is to help customers comparing smartphone cameras among each other and to DSLRs. We also hope that our method will guide smartphone makers in their developments and will thereby contribute to advancing mobile portrait photography.
The idea of contrast at a pixel, including contrast in colour or higher-dimensional image data, has traditionally been associated with the Structure Tensor, also named the di Zenzo matrix or Harris matrix. This 2 × 2 array encapsulates how colour-channel first-derivatives give rise to change in any spatial direction in x, y. The di Zenzo or Harris matrix Z has been put to use in several different applications. For one, the Spectral Edge method for image fusion uses Z for a putative colour image, along with the Z for higher-dimensional data, to produce an altered RGB image which properly has exactly the same Z as that of high-D data. So e.g. the contrast from RGB + NIR images can be fused such that Z in RGB takes on the same values as Z for 4-D data. As well, Z has been used as the foundation for the Harris interest-point or corner-point detector. However, a competing definition for Z is the 2 × 2 Hessian matrix, formed from second-derivative values rather than first derivatives. In this paper we develop a novel Z which in the first place utilizes the Harris Z, but then goes on to modify Z by adding some information from the Hessian. Moreover, here we consider an extension to a Hessian for colour or higher-D image data which treats colour channels not as simply to be added, but in a colour formulation that generates the Hessian from a colour vector. For image fusion, experiments are carried out on three datasets of 50 images each. Using the modified version of Z that includes Hessian information, results are shown to retain more details and also generate fused images that have smaller CIELAB errors from the original RGB. Using the new Z in corner-detection, the novel colour Hessian produces interest points that are more accurate, and as well generates fewer mistake points.
This paper presents multispectral imaging as an alternative to conventional color imaging that showed deficiencies. Thermal infrared images have useful signatures are insensitive to different illuminations and viewing directions. Multispectral imaging by the information fusion of the visible images and thermal infrared images provides rich data information that can be used in face recognition. Comparatively to traditional face recognition, multispectral imaging can separate illumination and reflectance information of facial images. The use of fusion of visible and thermal images in face recognition shows better performance than traditional imagery.
Realistic test data is needed to evaluate and rank the performance and quality of dehazing algorithms for image enhancement. Especially for professional photography and cinematography, this test data has to fulfill high quality standards including high dynamic range, sufficient resolution and natural color reproduction. For this purpose, we present a new multispectral data set that includes RGB and near-infrared (NIR) images captured by two professional digital motion picture cameras. Compared to existing data sets, the benefits of our set are threefold. Due to our two camera setting we are able to provide synchronous and well registered RGB/NIR image pairs captured at the same instant of time. High quality real image sequences allow future algorithms to take account of the temporal consistency of the dehazed output images. Furthermore, to facilitate a uniform and fair evaluation of different algorithms we provide ground truth images for selected RGB/NIR image pairs. The data set is freely available at http://www.arri.com/innovations/.