Research On Single-pixel Imaging Method Based On Radon Transform

Single-pixel imaging is a new type of computational imaging technology that uses structured light modulation to perform imaging based on the correlation between structured light and a single-pixel detector's measured value.Compared with array imaging technology,single-pixel imaging technology sacrifices time and cost in exchange for spatial information,reducing the detector's demand for resolution,especially in the invisible light band.The production cost of high-resolution array detectors is extremely expensive.Single-pixel detectors are cost-effective,and singlepixel imaging technology has become an effective alternative.Structured light is divided into non-orthogonal basis and deterministic orthogonal basis structured light.Random structured lights such as random speckles or binary matrices do not have orthogonality.Non-orthogonal basis,due to their crosstalk,require a huge number of samples to reconstruct high-quality images without considering the use of compressed sensing.Compressed sensing technology can reduce Sampling times,but reconstruction takes much time.Therefore,the traditional non-orthogonal substrate single-pixel imaging method has low imaging efficiency.The single-pixel imaging method based on Radon transform is called Radon single-pixel imaging method(RSI).Radon transform does not have orthogonality,and its performance in imaging quality is not as good as orthogonal basis single-pixel imaging method.The main work of this thesis is to study the theories,imaging capabilities and application of RSI.By modifying the Radon transform structured light basis,the image quality of RSI is improved.Radon transform could extract pixel coordinate information,and RSI imaging method can locate the target area in the scene according to the difference in the projection curve of the Radon transformation of the image in different directions.Only sampling and imaging the target area can improve the efficiency of single-pixel imaging.The main research work of this thesis is as follows:1.Introducing a variety of typical single-pixel imaging methods,establishing Radon single-pixel imaging method model,including post-modulation single-pixel imaging optical path,line scan Radon basis coding algorithm,single-pixel imaging method evaluation index.2.By analyzing the reconstruction error of RSI,approximate discrete Radon transform algorithm is used to modify the Radon transform base,reduces the error of the discrete Radon transform caused by Digital Micromirror array Device(DMD),and improves the imaging quality.Test the imaging ability of RSI under different impact factors(down-sampling,noise,quantization error,etc.).RSI can use down-sampling data to reconstruct clear images.The structural similarity coefficient of simple images can reach 0.98,and the structural similarity coefficient of complex images can reach0.91.Compared with other single-pixel imaging methods,RSI has significant resistance to additive noise.3.An adaptive Radon single-pixel imaging method is proposed.Based on the basic principle of Radon transform,the image's projection information in the horizontal and vertical directions is extracted to obtain the size and position of the target area in the scene.An adaptive Radon-Hadamard single-pixel imaging model is established,and only single-pixel sampling and imaging are performed on the target area.The number of samples is much lower than the number of samples for the whole scene,which reduces sampling time and improves the single-pixel imaging method's imaging efficiency.