P6 - Image Processing of Thermal Images using Wavelets
- Event
- SENSOR+TEST Conferences 2011
2011-06-07 - 2011-06-09
Nürnberg - Band
- Proceedings IRS² 2011
- Chapter
- IP - Poster Session
- Author(s)
- D. Rolle, H. Budzier, G. Gerlach - Technische Universität Dresden (Germany)
- Pages
- 123 - 123
- DOI
- 10.5162/irs11/ip6
- ISBN
- 978-3-9810993-9-3
- Price
- free
Abstract
This poster presentation covers novel results of image processing in infrared thermal imaging using wavelets instead of Fourier transformation. Thermal images often contain a plenty of noise to be extracted. The widely used discrete cosinus transformation (DTC) is not able to solve this problem because the noise signal does not depend on frequencies. In need of using small memories for large amounts of data there should also be taken benefit by using another technique of compression without the well-known artefacts of the JPEG-Standard at high compression rates. Furthermore it should be possible to amplify details of an image without changing the rest of the image. Goal of the work was to find a solution that not only localizes the image signal at the frequency space as the DCT does but also at the origin space. The wavelet transformation is able to localize the signal in both spaces at the same time as the uncertainty relation of Heisenberg allows it. The continious
transformation is defined by the direct product of the chosen wavelet function and the signal whereat the wavelet function is altered in order to match the translation and the frequency localization, respectively. The discrete wavelet transformation (DWT) can be obtained by convolving the discrete signal with the wavelet-filter.
The denoising is done by setting small coefficients to zero which is called thresholding. This simple method does not need any additional denoising or prediction techniques like Wiener-filter or others. Especially for image compression, the DWT is much more effective than the DCT. The reason is that the signal is composed by elements which vanish outside a defined region whereas cosinus bases never vanish. So there are less coefficients needed for describing the signal what can also be proven by the entropy of all coefficients in comparison to the DCT. The better entropy leads to a higher possible compression rate. Amplifying large coefficients gives the possibility to point out details and contrasts of the image. As the coefficients are localized in time and frequency, this method is effective without side effects the DCT would cause by amplifying the coefficients not localized in space. Denoising thermal images by DWT gives opportunities to image processing the DCT does not offer. At high compression rates the DWT causes a lower mean squared error than the DCT.