Optical Correlation Pattern Recognition And The Motion Blurred Image Restoration

The researches pursue the technologies of the optical correlation pattern recognition, the motion blurred image restoration, and the image denoising via wavelet transform. The research productions are suggested to be used in airborne laser weapon for recognition, tracking and indicating.The section of the Optical Correlation Pattern Recognition (OCPR) presents a new designed match filtering experiment using focus slant reference light, proposes a new iterative algorithm to synthesize the rotation invariant function, analyzes and testes the diffraction due to arrays of the pixels in TFT-LCD and the modulation mode of the TFT-LCD. The experiment of the hybrid Joint Transform Correlator (JTC) is also finished.The productions about OCPR include:①The developments and the applications of the OCPR are studied. The applications on OCPR in USA forces are given attention to. It is known that the OCPR has the properties of shift invariant and high speed because of parallel processing. It is used in military area such as numerous images pressing, high speed recognition, targeting et al. The integrated and miniature OCPR system costs less power and volume. It can work in gear from the vibration, dust, and background light. Vander Lugt Correlator (VLC) and Joint Transform Correlator (JTC) are the classic optical information processing systems. VLC and JTC do not have the properties of rotation or scale invariant. The methods to keep the OCPR system to be robust include the Mellin transform, the circular harmonic expansion, the synthetic discriminant function, MACH method, neural network,et al..②The match filtering experiment using focus slant reference light is designed and finished independently. It needs only one Fourier lens instead of two as in VLC. The optical distance in the new experiment is less than that in VLC. So the new OCPR system is compact and succinctness. The theory analysis about the new experiment is given. Good results are approached when a model of tank or a star-shape cooperative image is used as the target. The star-shape cooperative image has the properties of rotation and scale invariant itself. It could be used in autonomous navigational equipment for spacecraft or plane blind docking.③A new iterative algorithm to synthesize the rotation invariant discriminant function isproposed. The synthetic discriminant function via the new method has high discriminant ratio with rotation invariant when it is used in binary deduced JTC.④To analyze and teste the diffraction due to arrays of the pixels in TFT-LCD and the modulation mode of the TFT-LCD. The experiment of the hybrid Joint Transform Correlator (JTC) is also finished.The matrix of TFT-LCD is constructed with a metal mask that covers the inter-pixel spaces, obscuring the thin film transistors and preventing light leakage through the unswitched parts of the panel. It result in diffraction. The overlap of adjacent order of diffraction limited the spatial frequency used in optical information processing with TFT-LCD. The analysis shows that the relative amplitude spatial distributions of different diffraction orders are same.The positive optical image, the edge enhanced optical image and the negative optical image are obtained by changing the direction of the polarizer or analyzer when the spatial light is modulated by TFT-LCD with the digital images inputted.The researches on the motion blurred image restoration meet the needs of the airborne laser weapon system. The airborne optical system for long distance scouting or targeting or tracking often has a small range of view with high resolution. The motion including flying and vibration due to the engine and the airflow from the scramjet of the chemistry laser would like to blur the image terribly. It is necessary to restore the legible image from the blurred one to identify the weakest part of the enemy target.The section of the motion blurred image restoration includes:①The motion blur point spread function is analyzed. The study shows that it is a sparse matrix whose elements are nonzero alone and besides the diagonal.②Multiply new methods of the identification of the motion blur direction from the motion blurred image are proposed: (a)the motion blur direction identify method via 3×3 derivative operator.(b)the identification method of the motion blur direction from the motion blurred image via directional derivation using bilinear or C spline interpolation, (c)the identification method of the motion blur direction via Laplacian operator.The experimental results of (b) show that the bilinear interpolation needs less CPU time and the C spline method can achieves higher identification precision. To reduce the error caused by random, the combined method of weighted averages and bilinear and C spline interpolation is analyzed. The method via Laplacian operator can achieve high precision, but itwork only when the blur extent is large.It is the first time to analyze the identification error. The comparison between our new methods and Yitzhaky's method is illustrated, and the experimental results show that our new methods can achieves higher identification precision resisting noise, some drawbacks arisen in Yitzhaky's method are also overcome. Our new methods are valid for uniform motion, accelerated motion and vibration et al..The data experiments of identifying the motion blur direction show that the larger blur extent results in the higher precision of the motion blur direction identification.The characteristics of the original image also have influence on the error of the motion blur direction identification. Higher precision result tends to be achieved when the original image is isotropic and has more detail.③ As the motion blur direction identified, a new method of the identification of the blur extent is proposed. It is illumined from the theory of the joint transform correlator in optical information processing. The experimental results show that it can achieves high identification precision resisting noise. It is valid for uniform motion, accelerated motion and vibration et al..The motion blur direction and the blur extent can be identified from the motion blur image. Then the existence region of the point spread function can be educed, which is very important for the fast constringency of the iteration in the blind devolution. As the motion blurred direction identified, it can be rotated to the horizontal. Then the image restoration changes to be a question of one dimension from 2D. That means the image restoration could be parallel computed.The denoising step before the deconvolution is helpful to improve the restored image if the motion blurred image is dirtied with noises.The section of the wavelet denoising studies the real wavelet denoising and the complex wavelet denoising. The research on "Dual tree complex wavelet transform(DT CWT) + Trivariate Shrinkage (Y1, Y2, Y3) " is emphasized. It is presented that the step of DT CWT denoising before deconvolution can improve the restored image.The section of the wavelet denoising includes:①The good qualities and the basic theory of the wavelet denoising are studied. Multi wavelet denoising methods are introduced briefly, including which are SUREShrink, BayesShrink, AdaptBayesShrink, LAWML Shrink. The researches show that the complexwavelet transform should be canonized for denoising. The complex wavelet transform has the properties of approximate shift invariance and good directional selectivity, which is helpful to protect the detail of the image from being blurred in denoising. The dual tree complex transform was developed by noting that approximate shift invariance can be achieved with a real dwt by doubling the sampling rate at each level of the tree. The DT CWT exhibits limited redundancy: 4:1 for 2-D.②Referring to the Bivariate Shrinkage, the Trivariate Shrinkage (Y1, Y2, Y3 )is deduced. Trivariate Shrinkage (Y1 , Y2, Y3) perform very well in denoising, especially when the original image is dirtied terribly by noise. It is a good ideal to combine the DT CWT and the Trivariate Shrinkage (Y1, Y2, Y3) for denoising.③The comparing experiments of the denoising between the real and complex wavelet transform is done. It shows that the CWT denosing does better in protecting the details of the image and achieves restored image with higher definiting power.The better choices to remove the Gauss white noises from the image are summarized as(a)Use the complex wavelet transform. (b)Shrink the wavelet coefficients with ratio instead of threshold.(c)Adaptative estimating the local variances σx . (d)The absolutevalues of the wavelet coefficients in and between levels are correlative. Use the information as more as you can. (e)The noises model is an very important information for the wavelet denoising. (f)Select the right wavelet function, (g)adaptative change the computing window size.④The Gauss white noises are removed from the motion blurred image by CWT denoising method in the data experiments. The comparison experiments announce that we can get much better restored image if the noised motion blurred image is denoised by the method of DT CWT+Trivariate Shrinkage (Y1, Y2, Y3) before the deconvolution.The comparison experiments also devote reveal that the noises are easier removed from the motion blurred image as the blur extent is larger. This is because of the motion blurred image is smoother as the blur extent is larger.