• No se han encontrado resultados

5. LOS WARRANTS

5.11 FUNDAMENTOS DEL VALOR DEL PRECIO DE LOS WARRANTS

Thresholding the inverse transform overleaf results in the dark areas around the blood vessel edges are highlighted along with the lamina pores. This will result in an error when analysing lamina pore characteristics, as blood vessel features will be mistaken for pores.

2.18 iii

Thresholded image of 2.18Ü overleaf.

9 à Blood vesseledge

4

2.18 iv

The outline of the blood vessels was drawn on the

original image, and superimposed on the thresholded image.

Blood vessel edge masked

2.1.4.4 Image restoration- Deconvolution

From the above examples, it can be seen that processing the images in the frequency domain allows an improved method of pore visualisation. However, it is possible to improve the quality of the ’raw’ images to try and increase the power of the processing technique.

Section 1.2.2.1 described the principle of confocal optics, whereby only light from the focal plane of the raster-scanning laser spot will be detected by the

photodetector, and any light reflected from different planes or back-scattered will be ignored. By rejecting the defocused light far fewer photons are detected, increasing the quantum-photon noise component in the raw data. This 'image noise' is in addition to noise generated by the detection system itself - shot noise from random fluctuations in the photon intensity and thermal electronic noise from the detector itself. Since the properties of these noise sources are often known, it is possible to compensate for them to some degree in the final image. Additionally, any imaging system is far from perfect, and image quality is

degraded by aberrations present in the imaging system, as well as those that are caused by the eyes optical system. Moreover, factors in the environment can prevent the optical system from optimum performance. Once the final image has been generated, it is theoretically possible to correct for optical aberrations present in the imaging path. This is termed as deconvolving the image. In order to deconvolve the image, we need to have accurate knowledge of the optical aberrations of the image acquiring system as well as the eyes optical system. The aberrations generated by an optical system can be measured by the point- spread function (PSF)(Fincham and Freeman 1980). It is possible to determine how the changing environment affects the performance of the optical system and apply methods to correct for them, a technique known as adaptive optics. Adaptive optics has been used in ophthalmic imaging to image the retinal cone mosaic (Miller, Williams et al. 1996; Liang and Williams 1997; Liang, Williams et al. 1997). However, for the purposes of this study, these systems are not used

deconvolution onto images where the blurring function of the imaging system is not known- this is known as a

blind

deconvolution. Most blind deconvolution algorithms are adaptive- the first iteration makes a guess of the PSF of the imaging system, and after that ‘moulds’ the next PSF on the results obtained with each preceding iteration.

Deconvolution techniques are very useful in the restoration of images from both the HRT and the Zeiss cSLO. The two imaging systems differ fundamentally in their image acquisition techniques as described earlier in the chapter. With the HRT, the operator has to set an initial focus on the neuroretinal rim, but there is an ‘ image quality control’ feature that will the guide the operator as to how to change the focus and depth setting to grab an optimal image series. In addition, rather than grabbing images at a single plane of focus, the HRT grabs 32 images that are and progressively deeper levels through the optic nerve head, giving a 3- dimensional reconstruction of the area. For the Zeiss cSLO, the operator

manually focuses on the optic nerve head, and the images grabbed are all at a single plane of focus.

In view of these differences, the method of deconvolving the Zeiss and HRT images is quite different.

Deconvolving the H R T image

Each confocal section of the HRT image series is relatively deblurred already- the very point of confocal optics is to reject out-of-focus light. However, this system does not reject all of the out-of-focus light, and as a result the image retains substantial axial smearing. In addition to this, due to the rejection of non- confocal light, fewer photons are detected so a significant quantum-noise

element may be present in the final image. A way of ridding this element is by using an averaging technique as discussed in section 2.1.1.2, but this technique cannot be applied to the HRT images, as each confocal image is at a different level within the optic nerve head. Therefore use of a 3-dimensional blind deconvolution to remove the effects of the above noise elements is warranted.

Deconvolving the Zeiss image

Deconvolution of the Zeiss averaged image is also possible, using a 2-D blind deconvolution. Although averaging images increases the signal to noise ratio (section 2.1.1.2), it is possible that deconvolving the image will enhance it further. The following figures illustrate the application of a 2 and 3-D blind deconvolution to the Zeiss and HRT images.

Figs 2.19: Image Restoration I: Blind Deconvolution of HRT