Removal of Streaking Artefact in Images of the Pierre Auger Observatory Infra Red Cameras. Anna Anzalone^, Francesco Isgrò*, Domenico Tegolo

Transcription

1 Removal of Streaking Artefact in Images of the Pierre Auger Observatory Infra Red Cameras Anna Anzalone^, Francesco Isgrò*, Domenico Tegolo for the Pierre Auger Collaboration ^INAF Istituto di Fisica Cosmica e Astrofisica Palermo, Italy * Dipart. di Scienze Fisiche, Università degli Studi Federico II Napoli, Italy Dipart. Di Matematica e Informatica Università degli Studi di Palermo, Italy CRIS'10 Catania, Italy September 2010

2 Cloud Monitoring Clouds monitored also by infrared cameras Infrared images of the Field of View of the telescopes and of the cloud conditions around the site

3 Streaking Artefact COIHUECO Semi circular stripes are due to the operating hardware of the IR cameras It is induced by a misalignment of the chopper wheel edge during the scene acquisition phase Infrared images are currently analysed to produce both in automatic and supervised way, masks of cloudiness that provide information if each pixel in the telescope field of views, is contaminated or not by clouds. Streaking effect may interfere with the automatic cloud detection process

4 Method Sketch The method consists of 5 different parts and it's based on the application of a modified version of a combined multiscale Wavelet Fourier filter* : 1. Warping of the semi circular streaks into horizontal stripes (cartesian to polar plane) 2. Decomposition of the 2 D discrete signal of the image in a combination of wavelet functions at decreasing spatial resolution scale 3. Filtering in the Fourier frequency domain of the bands holding information of the horizontal structure presence 4. Inverse wavelet transform 5. Image representation in the original coordinates *Munch,Trtik, Marone,Stampanoni: 'Stripe and Ring Artefact Removal with Combined Wavelet Fourier Filtering','09

5 Artefact Characteristics Frame 1 Frame 4 Observing a big amount of non cloudy frames corrupted by the same artefact... Frame 2... Frame 3 Frame n... can be noted that they appear as : Alternating dark and light stripes Shaped as semi arches of concentric rings Located at nearly the same position in consecutive images

6 Artefact Characteristics Their characteristic intensity eases the extraction of the edges Their shape and location suggest to find the centre of the 'rings partially included in the image...

7 Linear Edge Extraction Points detected as 'edge' points in the most amount of frames are selected to evaluate the centre coordinates of the concentric 'rings' by the Generalized Hough Transform

8 Semi circular Stripes as Horizontal Stripes The image is warped into polar coordinate plane where all the 'semi rings' are mapped to horizontal stripes

9 Multiscale Wavelet decomposition of the image Purpose of this decomposition is splitting the structural information in horizontal,vertical and diagonal detail bands at decreasing spatial resolution scale Low pass filter and high pass filter are iteratively applied to the image in both vertical (along the image columns) and horizontal (along the rows) directions Horizontal, vertical, diagonal edges are highlighted by the high frequencies bands Whereas low frequency bands still retain all the most meaningful information The low frequency part is again decomposed by a low pass filter and a high pass filter, each time at a lower spatial scale and differently sized structure are detected Decomposition Depth is related to the maximum width of the stripes

10 Multiscale Decomposition Lev7... Lev 4 Lev 3 Lev 2 Lev 1 Symlet 16 Wavelet function High pass filter Scaling function Low pass filter

11 Fourier Transform and Filtered Wavelet decomposition FFT transform plus filtering of the horizontal detail bands Horizontal detail bands result 'cleaned' at different level of the decomposition

12 Results Original Filtered

13 Results Original Filtered

14 Structural features and quantitative values of the image are preserved Non Cloudy Image Filtered Image Original Image Difference Image The difference between the original image and the filtered one, shows that the only part of the image that was filtered out is related to the stripe structure

15 Structural features and quantitative values of the image are preserved Cloudy Image Filtered Image Original Image Difference Image No evidence of the meaningful structures of the original image

16 Intensity Vertical Profile by Columns Filtered Original

17 Intensity Vertical Profile by Columns Non Cloudy Image

18 Intensity Vertical Profile by Columns Filtered Original

19 Intensity Vertical Profile by Columns

20 Intensity Vertical Profile by Columns Filtered Original

21 NMSE as total Energy Loss NMSE Image index

22 De streaking in absence of strongly visible artefact NMSE = 2.65 this value can be justified by observing the plot of the vertical profile mean where discrepancies are present in the intensity levels but not in the shape of the vertical profiles

23 Filtered Original

24 Synthetic Artefact

25 Synthetic Artefact NMSE = 0.88 Filtered Original

26 Cloud detection based on Edge detection Here just a very rough example of the reduction of the amount of edges to deal with Edges before filtering Edges after filtering

27 Conclusions Results show high preservation of all meaningful information contained in the original image in different cloudy conditions The method reduces the necessity for a human operator to correct the artefact manually It can be still Improved to manage all the big amount of different type of cloudy images