Gradient in spectral and color images: from the Di Zenzo initial construction to a generic proposition

Abstract

In this paper, we define a generic gradient for color and spectral images, in regards to a proposed taxonomy of the state of the art. The full-vector gradient, taking into account the sensor's characteristics, respects the metrological properties of genericity, robustness and reproducibility. We construct a protocol to compare gradients from different sensors. The comparison is developed simulating sensors using their spectral characteristics. We develop three experiments using this protocol. The first one shows the results consistency for similar sensors. The second one demonstrates the genericity of the approach, adapted to any kind of imaging sensors. The last one focuses on the channel inter-correlation considering sensors as in the color vision deficiency case.

Ghost: Gradient Histogram of Spectral Texture

“…which is really the Jacobian matrix. The spatial correlation matrix, M is defined using full vector gradient (FVG) [6]:…”

“…Alternatively, one may choose to measure texture in terms of "cornerlikeness" using λ − [11]. If both edge-likeness and cornerlikeness are considered, then one could choose λ + + λ − [6].…”

“…In this work, we compute GHOST using a 3 × 3 filter constructed using a Gaussian function and its derivative as defined in [6]. We use the same reference as the one used in [13].…”