Well-Distributed Feature Extraction for Image Registration Using Histogram Matching

Abstract

Image registration is a spatial alignment of corresponding images of the same scene acquired from different views, sensors, and time intervals. Especially, satellite image registration is a challenging task due to the high resolution of images. In addition, demands for high resolution satellite imagery are increased for more detailed and precise information in land planning, urban planning, and Earth observation. Commonly, feature-based methods are applied for image registration. In these methods, first control or key points are detected using feature detector such as scale-invariant feature transform (SIFT). The numbers and the distribution of these control points are important for the remaining steps of registration. These methods provide reasonable performance; however, they suffer from high computational cost and irregular distribution of control points. To overcome these limitations, we propose an area-based registration method using histogram matching and zero mean normalized cross-correlation (ZNCC). In multi-spectral satellite images, first, different spectral responses are adjusted by using histogram matching. Then, ZNCC is utilized to extract well-distributed control points. In addition, fast Fourier transform (FFT) and block-wise processing are applied to reduce the computational cost. The proposed method is evaluated through various input datasets. The results demonstrate its efficacy and accuracy in image registration.

Research on 3D reconstruction technology of oblique images of closed coal yards

“…To improve the efficiency and accuracy of image matching, Mohsen Norouzi et al [12] combined the GGF algorithm with scale invariant features, calculated gradients using average ratios, and finally achieved FSC algorithm feature matching. Muhammad Tariq Mahmood et al [13] used cross-correlation methods between histogram and zero mean normalization for feature extraction and matching. UAV cameras can also be used for feature recognition and matching of fire images, and Shao Zhenlu et al [14] determined the range and severity of fires based on the temperature characteristics of thermal infrared images.…”