Edge Detection in Hyperspectral Imaging: Multivariate Statistical Approaches

Abstract

Abstract. Edge detection is well developed area of image analysis. Many various kinds of techniques were designed for one-channel images. Also, a considerable attention was paid to edge detection in color, multispectral, and hyperspectral images. However, there are still many open issues in edge detection in multichannel images. For example, even the definition of multichannel edge is rather empirical and is not well established. In this paper statistical pattern recognition methodology is used to approach the problem of edge detection by considering image pixels as points in a multidimensional feature space. Appropriate multivariate techniques are used to retrieve information which can be useful for edge detection. The proposed approaches were tested on the real-world data.

A fast multi-channel edge detection algorithm for vision-based autonomous spacecraft docking

“…Instead, more advanced and robust methods are preferred. Verzakov et al [17] proposed a multivariate statistical approach to multi-spectral edge detection. However, this approach does not employ any robust statistical techniques and the generated edge maps are sensitive to non-additive Gaussian noise.…”

An Efficient Hybrid Classification System for High Resolution Remote Sensor Data

“…Though edge detection in gray valued images is well studied, the task is less well defined in multi-or hyperspectral images. Techniques based on manifold learning [7], clustering and multivariate statistical approaches [17] were proposed to extract edges. Many of these techniques though perform satisfactorily on multispectral images, are not well suited for higher dimensional hyperspectral images.…”

An Enhanced Approach for Image Edge Detection Using Histogram Equalization (BBHE) and Bacterial Foraging Optimization (BFO)

“…[3], [4], statistics [5], mathematical morphology [6], [7]. Some more approaches for this are machine learning (ML) which is also analogous to computational statics [8], phase congruency and local energy [9], [10], multi resolution [11], a local feature's combination [12], and optimization strategies which are established on frequency models also called as phase congruency and the groups of pixels [13].The above approaches are generally connected with some of the preprocessing techniques like Gaussian filtering [14], [15]. Many methods were reported for higher quality of an image consisting of local contrast stretching [16], [17], graphical representation such as histogram equalization [18], [19], [20], contrast limited adaptive histogram equalization (CLAHE) [21], Bi histogram equalization (BHE) [22] which are spatial domains and DWT [23], [24], DCT [25] are commonly used compressed domain methods.…”