Deconvolution apparatus and method using a local signal-to-noise ratio

1. A method for a deconvolution of a digital input image (I(x i )) having a plurality of input voxels (x i ), in particular a digital input image obtained from a medical observation device, such as a microscope or endoscope and/or using fluorescence, the method comprising:

computing a local signal-to-noise ratio (SNR(x i )) within an input region (R(x i )) of the digital input image, the input region consisting of a subset of the plurality of input voxels of the digital input image and surrounding the current input voxel, and

computing a noise component (β(SNR)) from the local signal-to-noise ratio, the noise component representing image noise ( ([h*f](x i ), n(x i )) in the deconvolution,

wherein the noise component is limited to a predetermined minimum noise value (β min ) for a local signal-to-noise ratio above a predetermined upper SNR threshold value (SNR max ) and is limited to a predetermined maximum noise value (β max ) for a local signal-to-noise ratio below a predetermined lower SNR threshold value (SNR min ).

2. The method according to claim 1 , wherein the step of computing the local signal-to-noise ratio (SNR) includes computing a local signal level (S(x i )) in the input region (R(x i )) and a local noise level (N(x i )) in the input region (R(x i )), and wherein the step of computing the noise level includes computing at least one summed-area table for at least one contiguous region of input voxels (x i ).

3. The method according to claim 2 , wherein the step of computing the at least one summed-area table includes the step of computing the at least one summed-area table in a parallel manner using a prefix sum.

4. The method according to claim 2 , wherein computing the local noise level (N) includes computing a variance of the digital input image (I(x i )) using at least one summed-area table.

5. The method according to claim 4 , wherein the variance is computed after applying a linear derivative operator ( ) to the input image data (I(x i )) to obtain derivative image data (I′(x i )).

6. The method according to claim 5 , wherein the linear derivative operator ( ) is a gradient operator or an edge-detection filter.

7. The method according to claim 1 , wherein the step of computing the local signal-to-noise ratio (SNR(x i )) includes the step of computing a signal level (S(x i )) at an input voxel (x i ) in the input region (R(x i )), and wherein the step of computing the signal level includes convolving the input image data (I(x i )) with a blur kernel (k b (x i )).

8. The method according to claim 1 , wherein the noise component (β(SNR(x i ))) is computed as having a gradient

(

d

⁢

β

⁡

(

SNR

)

dSNR

)

,

which is smaller at the predetermined upper SNR threshold value (SNR max ) and/or at the predetermined lower SNR threshold value (SNR min ) than between the predetermined upper SNR threshold value (SNR max ) and the predetermined lower SNR threshold value (SNR min ).

9. The method according to claim 8 , wherein the step of computing the noise component (β(SNR)) includes the step of computing a trigonometric function.

10. The method according to claim 1 , wherein the deconvolution is a Lucy-Richardson deconvolution.

11. A non-transitory computer readable medium storing a program causing a computer to execute the image processing method according to claim 1 .

12. A medical observation apparatus, such as a microscope or endoscope comprising an image processor, the image processor being configured to carry out the method according to claim 1 .

13. An image processor for a medical observation apparatus, such as a microscope or endoscope, the image processor comprising:

a memory configured to store a digital input image (I(x i )) comprising a plurality of input voxels (x i ), and

a deconvolution engine configured to compute a deconvolved output image (f(x i )) from the plurality of input voxels,

wherein the deconvolution engine comprises a noise component (β(SNR(x i ))) which depends on a local signal-to-noise ratio (SNR(x i )) at an input voxel, the local signal-to-noise ratio being computed only in an input region (R(x i )) consisting of a subset of the plurality of input voxels of the digital input image,

wherein the image processor contains a predetermined upper SNR threshold value (SNR max ) and a predetermined lower signal-to-noise threshold value (SNR max ), and

wherein the noise component is limited to a predetermined minimum noise value (β min ) for a local signal-to-noise ration above the predetermined upper SNR threshold value (SNR max ) and to a predetermined maximum value (β max ) for a local signal-to-noise ratio below the predetermined lower signal-to-noise threshold value (SNR min ).

14. The image processor according to claim 13 , wherein the image processor comprises a summed-area-table generator, the summed-area-table generator being configured to compute a summed-area table of the digital input image and to compute the local noise level using the summed-area table.

15. A microscope or endoscope comprising the image processor according to claim 13 .