An automated system for the registration and comparison of photographic images in medicine

Abstract

A system is presented for digitization and automated comparison of photographic images of patients obtained at different times using a high-precision video camera. The images can be acquired either directly or from slides. The two images to be compared are registered using a complex geometrical and gray-level registration model including six parameters (planar, translation, rotation, magnification, linear transformation of the gray levels). The values of the registration parameters are automatically calculated by maximizing an integer similarity measure selected for robustness. The optimization of this function with respect to the registration parameters is performed using an adaptive random search strategy. The analysis of the differences between the registered images can be carried out through visual inspection of the subtraction image in which artifacts due to remaining infrapixel shifts have been suppressed.

<title>Weighted least squares for point-based registration in digital subtraction angiography (DSA)</title>

“…The standard form ofeq. (7) is a11x + 2a12x1x2 + a22x + 2a01x1 + 2a02x2 + a = 0 (8) which leads to the equivalent matrix notation XTAX + 2aTx +a00 = 0, To eliminate the second term in eq. (8) the coordinate system (reference system) must be aligned with the local coordinate system of the ellipse, given by its principal axes, using principal components analysis.…”

New hybrid stochastic-deterministic technique for fast registration of dermatological images

“…introduce differences in the images that need to be rectified. The parameters that describe the different acquisition conditions axe (VENOT, 1988) • a translation factor Tx, in the x-direction, that represents the relative translation of the projection of the object to the representation plane of the camera to the x-axis direction • a translation factor Ty, in the y-direction, that represents the relative translation of the projection of the object to the representation plane of the camera to the y-axis direction • a rotation factor R that represents the relative rotational displacement of the projection of the object to the representation plane of the camera • a scaling factor H caused by the varying distance of the object from the camera • a multiplicative normalisation factor NF and an additive background factor BG, both dealing with grey-level changes in the images.…”

“…Because of the increased incidence of malignant melanoma, researchers axe emphasising the need for automated melanoma recognition using image processing techniques (GANSTER et al, 2001;SCHMID-SAUGEON et al, 2003). Image registration is a significant issue in automated melanoma recognition, as, in many cases, there is a need to compare dermatological images taken at different time instances (VENOT et al, 1988). The numerous publications in the field of image registration applications in skin images WHITE and PEREDNIA, 1992;MCGREGOR, 1998;RONING and RIECH, 1998;Hsu et al, 1999) demonstrate the significant importance scientists attach to in these techniques.…”

Image registration of sequential transparent photographs to localize and detect new versus recurrent tumors in dermatologic and Mohs micrographic surgery

“…2,3,[5][6][7][8] The majority of dermatologic surgeons utilize standard and/or digital photographs to identify and compare skin lesions. This simple technique of transparent image overlay with image registration allows precise localization and the most accurate information possible regarding visual tumor localization.…”

“…These include the relative translation of the projection of the object to the representation plane of the camera in the x and y directions, a rotation factor, a scaling factor, and lighting and background variances. 5,6 Spatial transformation to achieve accurate image registration is based on correspondence. 3 Structural correspondence based on anatomic features allows relevant comparison of multiple images.…”