Method for generating super-resolution images having improved image resolution and measuring device
In the case of a measuring device for recording a sequence of individual images ( 6, 7, 8 ) in a non-visible spectral range, a method for generating an SR image ( 11 ) having an image resolution that is higher than an image resolution of the individual images ( 6, 7, 8 ) is provided, wherein, for the individual images ( 6, 7, 8 ), a displacement vector field ( 11, 12 ) is determined with a calculation of the optical flow and the individual images ( 6, 7, 8 ) are segmented into segments ( 20, 21, 22, 23, 24, 25 ) with regard to the values of the displacement vector field ( 11, 12 ), and an optimization method is carried out for calculating the SR image ( 11 ) from the individual images ( 6, 7, 8 ) with variation parameters individually assigned to the segments ( 20, 21, 22, 23, 24, 25 ).
1. A method for generating SR images ( 11 ) having improved image resolution, comprising: recording a sequence of individual images ( 6 , 7 , 8 ) comprised of pixels ( 9 , 10 ) in a non-visible spectral range and combining the individual images ( 6 , 7 , 8 ) in an optimization method to form an SR image ( 11 ) such that the individual images ( 6 , 7 , 8 ) of the sequence are modeled as samples of the SR image ( 11 ) by sampling functions, wherein an energy function is optimized in the optimization method, the energy function rates a deviation of the individual images ( 6 , 7 , 8 ) from the SR image ( 11 ) imaged by a mathematical model describing imaging properties of a measuring device ( 1 ) used for recording the individual images ( 6 , 7 , 8 ), the optimization comprises in a first step, calculating an optical flow with respect to the individual images ( 6 , 7 , 8 ) of the sequence automatically and assigning a displacement vector field ( 11 , 12 ) describing the optical flow to one of the individual images ( 6 , 7 , 8 ), in a second step, segmenting the individual images ( 6 , 7 ) into at least two segments ( 20 , 21 , 22 , 23 , 24 , 25 ), wherein the pixels ( 9 , 10 ) within one of the segments ( 20 , 21 , 22 , 23 , 24 , 25 ) in each case have a displacement vector ( 14 , 15 , 16 , 17 , 18 , 19 ) in a predefined sub-range of a value range of the displacement vector field ( 11 , 12 ), and in a third step, performing the optimization method, wherein in each case at least one separate variation parameter is varied for the segments ( 20 , 21 , 22 , 23 , 24 , 25 ), a subdivision of the SR image ( 11 ) into SR segments ( 35 , 36 , 37 ) is derived in each case from the subdivisions of the individual images ( 6 , 7 , 8 ) into the segments ( 20 , 21 , 22 , 23 , 24 , 25 ), and the sampling function or point spread function ( 38 , 39 , 40 , 41 , 42 , 43 , 44 ) assigned to one of the segments ( 20 , 21 , 22 , 23 , 24 , 25 ) of an individual one of the images ( 6 , 7 , 8 ) is applied to one of the SR segments ( 20 , 21 , 22 , 23 , 24 , 25 ) of the SR image ( 35 , 36 , 37 ) derived from the segment.
2. The method as claimed in claim 1 , wherein in the third step, for each of the segments ( 20 , 21 , 22 , 23 , 24 , 25 ) a displacement vector ( 29 , 30 , 31 , 32 , 33 , 34 ) is selected from the associated sub-range and is used for defining a start point of the optimization method.
3. The method as claimed in claim 1 , wherein an individual one of the images ( 6 , 7 , 8 ) of the sequence is selected as a reference individual image ( 7 ), or the displacement vector fields ( 11 , 12 ) are calculated with regard to the reference individual image ( 7 ).
4. The method as claimed in claim 1 , wherein before the calculation of the optical flow, in a global registration for each individual one of the images ( 6 , 7 , 8 ), a global or rigid transformation is calculated with which corresponding image contents of the individual images ( 6 , 7 , 8 ) can optimally be brought to congruence with one another or with a reference individual image ( 7 ).
5. The method as claimed in claim 1 , wherein the energy function has a term which models an image noise.
6. The method as claimed in claim 1 , wherein the variation parameters in each case describe a relative image displacement of the segment ( 20 , 21 , 22 , 23 , 24 , 25 ), or the variation parameters in each case define a sampling function or a point spread function ( 38 , 39 , 40 , 41 , 42 , 43 , 44 ).
7. The method as claimed in claim 1 , wherein in the optimization method, after optimization of the variation parameters, an adapted SR image ( 11 ) is created iteratively, for which the variation parameters are optimized again.
8. The method as claimed in claim 1 , wherein the third step involves defining for which of the segments ( 20 , 21 , 22 , 23 , 24 , 25 ) of the individual images ( 6 , 7 , 8 ) the optimization method is performed, or the optimization method is performed only for those of the segments ( 20 , 21 , 22 , 23 , 24 , 25 ) of the individual images ( 6 , 7 , 8 ) or SR segments ( 35 , 36 , 37 ) of the SR image ( 11 ) whose associated displacement vectors ( 14 , 15 , 16 , 17 , 18 , 19 ) lie in a predefined range or outside a predefined range.
9. A measuring device ( 1 ), comprising an image recording device ( 2 ) for recording a sequence of individual images ( 6 , 7 , 8 ) in a non-visible spectral range, and an image evaluation unit ( 5 ) configured and set up for performing a method as claimed in claim 1 .