Motion Estimation from a Single Blurred Image

Transcription

1 Motion Estimation from a Single Blurred Image Image Restoration: De-Blurring Build a Blur Map Adapt Existing De-blurring Techniques to real blurred images

2 Analysis, Reconstruction and 3D reconstruction from a single blurred image Example More info: giacomo.boracchi@polimi.it

3 Blurred Image Analysis and Restoration Restoration of Radial Blur

4 Blurred Image Analysis and Restoration Restoration of Radial blur

5 Blurred Image Analysis and Restoration Rotational Blur Estimation From a single Image

6 Blurred Image Analysis and Restoration Rotational Blur Estimation From a single Image

7 Blurred Image Analysis and Restoration Rotational Blur Estimation From a single Image Blur is Space Variant Estimation of 3d camera rotation axis Angular speed From a single image

8 Blurred Image Analysis and Restoration And then. Blur Removal Denoising! i.e. a restoration algorithm

9 Blurred Image Analysis and Restoration Restoration of Rotational Blur

10 Blurred Image Analysis and Restoration Restoration of Rotational Blur

11 Blur Analysis and Restoration -- Projects Implement a Rotational Blur Removal algorithm that takes into account also noise. Deblurring on spherical surfaces More info: giacomo.boracchi@polimi.it

12 Blur classification in a single blurred image Given a Blurred Image, determine if the image is blurred and in that case, if the blur is due to object motion camera shake out of focus Depth of field These information are necessary for image restoration

13 Depth of Field Blurred Object Segmentation

14 Depth of Field Blurred Object Segmentation

15 Depth of Field Blurred Object Segmentation

16 Blur classification and Segmentation -- Projects Improve the segmentation algorithm, exploiting image priors Embed the blur estimation algoirthm Restoration: blur inversion + noise removal Blur Classification More info: giacomo.boracchi@polimi.it

17 Low-light Image Restoration Taking satisfactory pictures at low-light conditions is challenging. Pictures acquired with a short exposure-time have low SNR and are very noisy because of a high gain (ISO number).

18 Low-light Image Restoration

19 Low-light Image Restoration

20 Introduction Taking satisfactory pictures at low-light conditions is challenging. Pictures acquired with a short exposure-time have low SNR and are very noisy because of a high gain (ISO number). Typically the exposure time is increased in order to improve the SNR of the acquired image. But this also increases the risk of blur, because of movements occurring in the extended exposure.

21 Low-light Image Restoration

22 Low-light Image Restoration A variety of solutions: Lenses Stabilization Different Acquisition Strategies In particular [Tico06] and [Yuan07] proposed two methods that use differently exposed images one with a long exposure time (blurred but with negligible noise) one with a short exposure time (noisy but with negligible blur) The noisy image is used to estimate the blur PSF allowing to restore the blurred image (deblurring) [Tico06] Tico, M., "Estimation of motion blur point spread function from differently exposed image frames," Proc. 14th Eur. Signal Process. Conf., EUSIPCO 2006, Florence, Italy, September 2006 [Yuan07] Yuan, L., J. Sun, L. Quan, and H.-Y. Shum, "Image deblurring with blurred/noisy image pairs," ACM Trans. Graph., vol. 26, no. 3, July 2007

23 Low-light Image Restoration Acquire a sequence of short exposure images (frames) and jointly denoise them, using a video denoising algorithm

24 BM3D Denoising More info:

25 BM3D Aggregation

26 BM3D Aggregation

27 BM3D Aggregation

28 Denoising vs Deblurring Long exposure, camera shaked image

29 Denoising vs Deblurring One of the short exposure, noisy image

30 Denoising vs Deblurring A detail from restored with Tico et al. algorithm Visible artifacts due to mismatches between assumed blur model (invariant PSF, linearity) and real blur.

31 Denoising vs Deblurring A detail from image restored with our algorithm There are less artifacts. Modeling is accurate. Denoising is less ill-posed than deblurring.

32 Denoising vs Deblurring A detail from image restored with Tico et al. algorithm

33 Denoising vs Deblurring A detail from image restored with our algorithm Not all details can be recovered by denoising because SNR is too low.

34 Denoising vs Deblurring A detail from the noisy image

35 Denoising vs Deblurring A detail from image restored with our algorithm Not all details can be recovered by denoising because SNR is too low.

36 Denoising vs Deblurring A detail from image restored with our algorithm Not all details can be recovered by denoising because SNR is too low. More info: giacomo.boracchi@polimi.it