Research On Optimization Of Coded Exposure Technology In Information Integration

The information integration of remote sensing imaging systems needs to efficiently integrate various sub-imaging systems that base on different platforms and utilize different wave bands, which requests these sub-systems to supply specious enough information. As a very important part of the entire remote sensing imaging systems, the image motion blur problem of the the visible remote sensing imaging system of unmanned aerial vehicles(UAV) will inevitably disturb the final decision. With the fast development of UAV, the visible remote sensing imaging system of UAV plays a more and more important role in the entire remote sensing imaging systems. Therefore, to solve the image motion blur problem for the visible remote sensing images of UAV is an ugent task.The image motion blur of the visible remote sensing images of UAV is mainly caused by the relative movement between the imaging device and the scene. There are mainly two ways to resolve the image motion blur problems. One is to use the motion deblurring methods in a post-processing. The other is to utilize high-speed imaging techniques. However, traditional image motion deblurring is an ill-posed problem and usually behaves badly when used for large-extent motion blurred images. High-speed imaging technique needs specialized hardware device and illumination, which has a very limited application scope and is often very expensive. Recently, a new technique called coded exposure which belonges to computational photography is proposed. Coded exposure technique creatively changes the ill-posed motion deblurring to a well-posed problem, which paves a new way for motion deblurring problem. Meanwhile, coded expsure can be seen as a random sample process along the temporal dimension, which is cater for the compressive sensing theory. Hence, from this view, the low frame-rate coded exposure photography has the possibility to achieve high-speed imaging.Aiming at obtaining clear images of fast moving object, this dissertation utilized coded exposure technique as a skeleton and made a deep study both on image motion deblurring and high-speed imaging problem. The main achievements and contributions of the dissertation are listed as follows:(1) An optimal binary sequences search method for coded exposure photography based on Legendre Sequence and Genetic Algorithm(GA) is poposed. The search space of the optimal binary sequence will grow exponentially with the increase of the binary sequence. The exsiting code search method can easily fail to find good long sequences. In this dissertation, we find the optimal binary sequences for coded exposure photography based on Legendre sequence which is a low aperiodic autocorrelation code used frequently in information theory commnity. We further improve the quality of the original Legendre seqences by rotating and appending operator and obtain the modified Legendre sequences. We then use the modified Legendre sequences as the initial popolation of GA, and use the crossover and mutation operations of GA to get the final result. The obtained binary sequences by the proposed method can satisfy the request of invertibility for coded exposure photography and can significantly reduce the noise of the motion deblurred image. In addition, since Legendre sequences can be computed very quickly by a close-form formula, the proposed method can generate good-quality sequences with very little computational load, which is especially suitable for searching long sequences.(2) An optimal code search method in the presence of photon noise is proposed. The process of coded exposure photography inevitably increase CCD noise. For practical purpose, we should consider the influence of noise when finding the optimal code for coded exposure photography. Firstly, the sensor noise in the process of imaging is systemically analyzed based on the affine noise model. Then, the basis of Raskar’s code is theoretically analyzed for the first time and based on the signal to noise ratio(SNR) gain of deblurred image this dissertation presents an analytic formula between the photon noise and the structure of the optimal code. Finally, based on the noise calibration of real coded exposure camera, this dissertation proposes new criteria for the optimal code and designs a fitness function and obtains the optimal code using the genetic algorithm. The proposed method significantly improves the efficience over the state-of-the-art code search method and meanwhile increase the signal to nosie ratio(SNR) of the motion deblurred image with the largest extent.(3) An automated motion blur identification method for coded exposure photography based on image power spectrum statistics is proposed. Coded exposure images have non-smooth blur, which makes the popular motion blur identification methods be impotent. We analyze the image power spectrum statistics and observe that the motion blur length has some relations with the residual sums of squares(RSS) of the image power spectrum statistics in a least squares sense. That is, the power spectrum statistics of the obtained deblurred image using the correct estimated blur length corresponds to the lowest value of the RSS. Giving an initial estimated blur length and using an iterative method, we model the power spectrum statistics of the deblurred images with polynomial fitness, and find the lowest RSS. Since the recovery of coded exposure image can use directly matrix inverse method which is very efficient, we can easily find the correct blur length using a global search method.(4) A high-speed video capture method using a single flutter shutter camera is proposed. Based on the consideration of easy achievement in modern sensors, this dissertation further exploits the possibility of the recovery of high-speed video(HSV) by a single flutter shutter camera. Taking into account different degrees of smoothness along the spatial and temporal dimensions of HSV, this dissertation proposes to use a three-dimensional hyperbolic wavelet basis based on Kronecker product to jointly model the spatial and temporal redundancy of HSV. Besides, we also utilize the flexibility of 1? minimization problem and incorporate the total variation of temporal correlations in HSV as a prior knowledge to further enhance our reconstruction quality. We recover the underlying HSV frames from the observed low-speed coded video by solving a convex minimization problem.