Warp and blur imaging model consistent with the three constraints of imaging through refractive turbulence

Abstract

Point spread function (PSF) for imaging through inhomogeneous refractive medium, such as atmospheric turbulence, is bounded by three constraints [Opt. Eng. 52, 046001 (2013)OPEGAR0091-328610.1117/1.OE.52.4.046001]. PSF is non-negative, band-limited, and the third constraint, related to the energy conservation principle, warrants the absence of fluctuations in the image of a uniformly bright object. We develop a version of the common warp and blur model for the anisoplanatic image distortions by turbulence that satisfies these three constraints. In order to comply with the third constraint, our model supplements warp and blur by a random PSF power, which was found to be related to the Jacobian of the warp field. We illustrate statistics of the warp and blur using a simple example of anisoplanatic phase screen.

Content-Seam-Preserving Multi-Alignment Network for Visual-Sensor-Based Image Stitching

“…To solve the model inadequacy of global alignment and further improve image deformation quality, some spatially varying warping methods have been designed recent years [20][21][22][23][24][25]. As a pioneering work, Zaragoza et al [10] proposed an as-projective-aspossible method, where a moving direct linear transform (DLT) method was designed to adjust the projective warp.…”

“…For instance, the blue people in the overlapping regions show serious ghosting in the second line of Figure 4b. In contrast, the NISwGSP method [23] preserves the structure of the overlapping regions by the integration of global similarity transformation, but it causes significant artifacts in the non-overlapping regions, such as the street lamp, which has an obvious inclination in the third line of Figure 4c. As given in Figure 4d, the salient objects in the overlapping regions are preserved by the REW method [24], but there is misalignment at the stitching seam in some cases.…”

“…In this section, we compare the proposed image-stitching method with six other methods, i.e., the APAP method [10], NISwGSP method [23], REW method [24], SPSO method [16], JVCIR method [39], and NC method [41]. The APAP method [10], NISwGSP method [23], REW method [24], and SPSO method [16] are traditional image-stitching methods, while the JVCIR method [39] and NC method [41] are deep image-stitching methods.…”

“…In this section, we compare the proposed image-stitching method with six other methods, i.e., the APAP method [10], NISwGSP method [23], REW method [24], SPSO method [16], JVCIR method [39], and NC method [41]. The APAP method [10], NISwGSP method [23], REW method [24], and SPSO method [16] are traditional image-stitching methods, while the JVCIR method [39] and NC method [41] are deep image-stitching methods. It is worth noting that the results of the APAP method [10], NISwGSP method [23], REW method [24], and JVCIR method [39] were obtained by running the public source code, and the results of the SPSO method [16] and NC method [41] were obtained by the implementation by us.…”

“…The APAP method [10], NISwGSP method [23], REW method [24], and SPSO method [16] are traditional image-stitching methods, while the JVCIR method [39] and NC method [41] are deep image-stitching methods. It is worth noting that the results of the APAP method [10], NISwGSP method [23], REW method [24], and JVCIR method [39] were obtained by running the public source code, and the results of the SPSO method [16] and NC method [41] were obtained by the implementation by us. For space limitations, only the image-stitching results of some typical scenes and the challenging scenes are presented in this section.…”