Passive three-dimensional defocusing method refers to finding the relative ambiguity in two defocusing images of an object, that is, the ratio of two defocusing parameters, and then obtaining the three-dimensional structure of the object through the relationship between the ratio of two defocusing parameters and the fuzzy parameters in the optical system. This method was first put forward in 2008. Because the three-dimensional defocus method must be based on the texture characteristics of the object surface when calculating the ambiguity, it is not easy to obtain the correct relative ambiguity when the texture characteristics of the object surface change greatly, which brings difficulties to the later calculation.

2. Restore shape and texture from light and shade, and restore stereoscopic vision of shape and luminosity;

The basic principles of these three methods are very simple, and they all need to obtain the relevant information of the surface of the object first, and then restore the shape of the object. Shape recovery from texture is to obtain the orientation of the object surface under certain conditions by using the characteristics that the texture of the object surface has different surface orientations in the imaging process.

The resolution of TOF method is relatively low, about 1 mm, but with the continuous development of optical holography and photon calculation, the resolution can reach micron level. The differences between the time-of-flight method and the trigonometric method are as follows: firstly, it is difficult for the time-of-flight method to change the measurement accuracy to adapt to different measurement ranges, which can be seen from the formula (it can be seen that the distance is a linear function of the time of flight, but the scale factor here, namely the flight speed, is not controlled by the measurement system. Secondly, the ranging accuracy is improved by the time accuracy of pulse signal measurement, and the spatial resolution of imaging spot position is improved by trigonometry, thus the ranging accuracy is obtained. This method is generally only used for large-scale absolute distance measurement, because it requires too high time resolution of the timing system, which brings great practical difficulties to the technical realization.

Moire profilometry mainly uses two gratings in the measurement system, the main grating and the reference grating. When they are superimposed together, moire fringes are formed, that is, the contour lines formed on the surface of objects. Moire profilometry can be divided into practical projection moire and shadow moire. Therefore, in order to make this technology more practical, people gradually combine moire method with A technology and computer image processing technology. After the subsequent phase shift technology is applied to moire method, phase shift moire appears, which solves the problem that the surface roughness can not be distinguished from a single moire image.

The measurement range of FTP method is limited, so in order to obtain the correct surface shape information, the stiffness change and phase change of the measured object must meet the following conditions:

By analyzing the full-field fringe, the phase value of each point can be obtained, because each point in the figure carries height information, so interpolation is no longer needed; Can automatically judge the unevenness of stripes; Easy to combine with computer to process results; Have certain ability to suppress noise.

SPD and PMP can realize automation in data processing without manual intervention, but they need to manually determine the position of the filter window, so there are certain limitations. Among them, the speed is faster than the other two methods, and it is less affected by non-sinusoidal fringes, so the machining program can be completed in the hardware system. However, if the surface of the object is defective, the phase jump will be too large, which will lead to a lot of errors.

MMP method can be used for vertical measurement without phase unwrapping. This method is suitable for objects with large height variation, scattered space and deep holes. When the object is in a dynamic process, the shape of the surface is constantly changing, which brings great difficulties to the vertical scanning of projection fringes at every moment, and it is impossible to achieve accurate measurement, so it is more suitable for the three-dimensional measurement of static objects.

The principle of Fourier transform profilometry: generally speaking, the fringe pattern is transformed from spatial domain to frequency domain. Only the fringe frequency is kept in the frequency domain, and the high-frequency noise and carrier are removed, and then the fringe pattern is restored from the frequency domain to the spatial domain by inverse Fourier transform, so that a fringe field distribution appears, which exists in the form of complex numbers, and then we can calculate the phase value of the fringe field by complex number operation.

The full name of DCT transform is discrete cosine transform, which is mainly used for data or image compression. It can transform the signal in spatial domain into frequency domain and has good decorrelation performance. DCT transform itself is lossless, but it creates good conditions for Havermann coding in the following fields, such as quantization and image coding. At the same time, because the DCT transform is symmetrical, the original image information can be restored at the receiving end by using the inverse DCT transform after quantization coding.

DTFT is a discrete-time Fourier transform, which is used to represent the frequency spectrum of continuous signals.

DFT:DFT is a discrete Fourier transform, which is aimed at discrete signals and frequency spectrum. DFT is a variant of DTFT, which actually changes the continuous time T into nT. Why are you doing this? Because the computer works in the digital environment, it can't see or process the continuous signal in reality, so it can only do discrete calculation, which is as close to the continuous signal as possible in reality. So DFT was created, so that we can use tools to analyze signals. Usually, we rarely have the opportunity to use DTFT directly.

FFT: First of all, DCT is a form of DFT. The so-called "cosine transform" means that in the expansion of DTFT Fourier series, if the expanded function is a real even function, its Fourier series only contains cosine terms, and then the cosine transform can be discretized (DFT), so it is called discrete cosine transform (DCT). Actually DCT belongs to a subset of DFT. DCT is widely used in speech and image processing.

Video 640x480, 30fps, YCbCr4:2:2, then the transmission rate requirement: 640x480x30x (8+8/2+8/2) =147.456 Mbps. The main sampling formats are YCbCr 4:2:0 and YCbCr 4:2:2. Among them, YCbCr 4: 1: 1 is commonly used, which means: every point saves an 8-8bit brightness value (that is, a Y value), and every 2×2 points saves a Cr and Cb value, so the naked eye's perception of the image will not change much. So the original RGB(R, G, B is 8-bit unsigned) model, each point needs 8x3=24 bits, but now it only needs 8+(8/4)+(8/4)= 12 bits, and each point occupies 12 bits on average. This will compress the image data by half.

YUV(YCbCr) sampling format

Spacing and width

Take a picture with a pixel of 640*480 and a color of 24 bits (3 bytes) as an example: Width: indicates the logical width of the picture, here is 640. This value has nothing to do with color depth and nothing to do with others. The width you see is its value pitch: it indicates the number of bytes or span of a line of data in the picture, which should be 640*3 here, because the width of the picture is 640.

Any periodic function can be regarded as the superposition of sine waves with different amplitudes and phases:

The rising part of all sine waves gradually steepens the curve that was originally slowly rising, while the falling part of all sine waves offsets the part that continues to rise when it rises to the highest point, making it a horizontal line. This adds a rectangle. Not just rectangles, but any waveform you can think of can be superimposed with sine waves in this way.

The black line in front is the sum of all sine waves, that is, it is more and more close to the figure of rectangular wave. Sinusoidal waves arranged in different colors are components of rectangular waves. These sine waves are arranged from front to back according to the frequency from low to high, and the amplitude of each wave is different. Careful readers must have found that there is a straight line between every two sine waves, which is not a dividing line, but a sine wave with an amplitude of 0! In other words, some sine wave components are unnecessary in order to form a special curve.

A frequency of 0 is also called the DC component. In the superposition of Fourier series, it only affects the whole waveform up or down relative to the number axis, without changing the shape of the waveform.

Frequency domain image, also called frequency spectrum, is:

It can be found that in the frequency spectrum, the amplitudes of even terms are all zero, which corresponds to the colored straight line in the figure and the sine wave with amplitude of zero.

Fourier transform is a more important but slightly complicated application-solving differential equations. I don't need to introduce the importance of differential equations too much. It is used in all walks of life. But solving differential equations is a troublesome thing. Because in addition to calculating addition, subtraction, multiplication and division, differential integral is also calculated. Fourier transform can make differential and integral become multiplication and division in frequency domain, and college mathematics instantly becomes elementary school arithmetic.

Since sine wave is periodic, we need to set something to mark the position of sine wave. Those in the picture are little red dot. Little red dot is the peak closest to the frequency axis. How far is this peak from the frequency axis? In order to see more clearly, we project the red dot to the lower plane, and the projection point is represented by a pink dot. Of course, these pink dots only represent the distance from the peak to the frequency axis, not the phase.

Time difference is not a phase difference. If all periods are regarded as 2Pi or 360 degrees, the phase difference is the proportion of the time difference within a period. We divide the time difference by the period and multiply it by 2Pi to get the phase difference.

In the complete stereogram, we divide the time difference obtained by projection by the period of frequency in turn to get the lowest phase spectrum. Therefore, look at the spectrum from the side and the phase spectrum from the bottom.

All phases except 0 in the phase spectrum are π. Because cos(t+Pi)=-cos(t), in fact, the wave with phase Pi just flips up and down. For the Fourier series of periodic square waves, such a phase spectrum is very simple. In addition, it is worth noting that because cos(t+2Pi)=cos(t), the phase difference is periodic, and Pi is the same as 3pi, 5pi and 7pi. The range of the artificially defined phase spectrum is (-π, π), so the phase difference in the figure is π.

Fourier series is a periodic continuous function in time domain and an aperiodic discrete function in frequency domain. Fourier transform is actually the Fourier transform of an infinite periodic function.

These discrete sine waves are getting closer and closer, and gradually become continuous ... The superposition of discrete spectra has become the accumulation of continuous spectra. Therefore, in calculation, it has changed from a summation symbol to an integral symbol.

A vertical imaginary axis: the real axis and the imaginary axis are isomorphic to form a complex plane, also known as the complex plane. A function of rotating multiplication of imaginary number I.

It is a point that moves in a circle on the complex plane with time, and becomes a helix on the time axis with time. If we only look at its real part, that is, the projection of the left helix, it is the most basic cosine function. The projection on the right is a sine function.

E^(it) can be understood as a spiral rotating counterclockwise, then E (-it) can be understood as a spiral rotating clockwise. Cos (t) is half of the superposition of these two spirals with different rotations, because the imaginary parts of these two spirals cancel each other out.

For the convenience of looking at the picture like a big conch, I only showed the positive frequency part, but not the negative frequency part. If you look carefully, you can see every spiral on the conch map clearly. Each spiral has a different amplitude (rotation radius), frequency (rotation period) and phase. Connecting all the spirals into a plane is this conch picture.

The two-dimensional Hanning filter window is used for digital weighted filtering, which gives greater weight to the fundamental frequency and less weight to the part far away from the fundamental frequency, making the phase unwrapping work easier.