Brief introduction of optical conversion processing

Using modern optical information processing technology is one of the important development directions of remote sensing image processing.

Optical information processing usually uses the two-dimensional Fourier transform capability of a thin lens to perform optical Fourier transform on a film image with coherent light (laser) or partially coherent light (white light) in an optical information processing system, then transforms the image in the spatial domain into information in the frequency domain, and then enhances the image by "modifying" spectral filtering. Therefore, the essence of optical information processing is optical transformation processing based on Fourier optical theory. Its principle is shown in Figure 4-6.

Figure 4-5 Schematic diagram of optical principle of light ratio image

Figure 4-6 Schematic diagram of optical conversion technology

In the figure, P 1 is the original figure, P2 is the Fourier spectrum generated by monochromatic light source S, and P3 is the inverse transformation of P2. If the spectrum of P2 is not modified, P3 is only the inverted image of P 1; When lenses with different transmission characteristics are set at P2, so that some spectra pass, while others do not, the image generated at P3 is a filtered image. According to the properties of the filter lens (Figure 4-7) and its enhancement effect, it can be divided into:

Figure 4-7 Spatial Filter

1. The low-pass filter is a circular hole (Figure 4-7a), which can block high-frequency components far away from the optical axis and let low-frequency components near the optical axis pass. It can stabilize the fine structure in the original image and enhance the main structure.

2. The Qualcomm filter (Figure 4-7b) only attenuates the zero frequency appropriately to weaken the background, highlight the high-frequency components with sudden change in optical density, enhance the edges and fine structures, and help to display hidden structures.

3. Band-pass filtering Many ground objects have specific spatial frequencies, so making corresponding band-pass filters (Figure 4-7c) only allows these frequencies to pass, thus achieving the purpose of enhancement.

4. Select a sector filter (Figure 4-7d 1) or a slit filter (Figure 4-7d2) for directional filtering, and only allow the spectrum perpendicular to the sector axis or slit direction to pass, so as to extract the linear information in this direction; Linear images in different directions can be enhanced by changing the direction of the central axis or slit of the sector. Therefore, it can be used as a linear structural reinforcement.

The above four kinds of optical information are transformed by changing the amplitude distribution on the spectral plane, so they are called amplitude filtering or simple spatial filtering. In addition, phase filters, grating filters, matched filters, etc. It can also be provided. Among them, grating filtering can not only enhance the linear structure in different directions, but also realize mathematical operations such as image addition and subtraction, and obtain various enhancement effects; Matched filtering is mainly used for image recognition and detection of specific target information.

In addition, white light (partially coherent light) is used instead of monochromatic light source, and color filter (halftone screen film-a grating that can convert the image density of black and white images on the object plane into spatial spectral distribution) is used instead. After optical Fourier transform, a color image encoded according to the original image density level is obtained on the image plane. This is also called optical coding, or optical pseudo color frequency coding, and it is also one of the commonly used optical transformation processing methods.

Optical information processing has the advantages of large capacity, two-dimensional parallel processing, fast image transformation, convolution and correlation, and has achieved good results in practical applications, showing great potential. But it also has its own limitations, for example, it is not as flexible as digital image processing, and the processing system itself cannot control, analyze and judge. Optical-mechanical hybrid processing combines the fast and large capacity of optical information processing with the flexibility and high precision of digital image processing, which will be the development direction of remote sensing image processing in the future.