Infrared Image De-noising Based On Multi-scale Geometric Transformationand Spiking Cortical Model

With the mature of infrared imaging technology and the emergence of infrared thermal imaging instrument which processes a low cost but high applicability, infrared image plays an important role in various departments of national economy. However, each procedure of the infrared imaging will cause noise pollution, which leads to the problems of low contrast and the blur between objectives and boundaries. Therefore, a de-noising algorithm to solve these problems appears particularly important. Multi-scale geometric transform used for image processing because its good time-frequency performance and ability to deal with multi-dimensional signal problems. Meanwhile, many scholars choose to apply spiking cortical model into image de-noising and enhancing, because the processing method of timed matrix in spiking cortical model can meet the visual characteristics of the human eyes. This dissertation focuses on the image de-noising algorithm based on multi-scale geometric transformations such as complex Contourlet transform, complex discrete Shearlet transform and spiking cortical model, The main field of application is infrared images, the main contribution of this dissertation are as follows:1. Color image de-noising based on improving SCMAccording to the blurring problem which present color image de-noising algorithm has, this dissertation uses the synchronous pulse burst property of spiking cortical model(SCM) and the characteristic that impulse noise is significantly different from surrounding image pixels to locate noises in each channel of color images separately, and then filter and remove noises in each channel according to the test results to keep non-noise pixels unchanged. In experiment, this method has a good color image de-noising result, and preserve image details in the image.2. Improved threshold and threshold functionFirst of all, based on the of defects of soft-threshold and hard threshold during the analysis of image de-noising, this dissertation enhances the threshold function to achieve better de-noising results; secondly, considering the fact that decomposition scale and sub-band will suffer a different degree of noise pollution after the image complex Contourlet transform and complex Shearlet transform, this dissertation also chooses the corresponding threshold value, therefore the “over killing ” phenomenon of unified value towards the coefficients of transform domain can be overcame. Moreover, it can realize a self-adaptive threshold of infrared image noise removal.3. Study on Infrared image de-noising algorithm based on multi-scale geometric transformation and SCM.Most of the existing multi-scale geometric transformation of image de-noising algorithms use the shrinkage of the transform domain to handle all the coefficients. The method is simple, but during the processing part, parts of non-noise pixels will also suffer the same shrinking dealing, which can damage the de-noising results. This dissertation using SCM to overcome this shortcoming, applies the combination of multi-scale geometric transformation and spiking cortical model to realize the image noise removal. Detect noise pixels by employing the synchronous pulse burst property of SCM in transform domain, then the noise pixels were processed by threshold shrinkage. Finally, noise can be effectively removed and non-noise pixels can be kept, and the image edge details remained. This experiment can be divided into two categories: one is the infrared image de-noising based on Contourlet and SCM, the other is image de-noising algorithm based on Shearlet and SCM. Each algorithm mentioned before will be conared by experiment one by one, from the subjective visual and objective data analysis of the experimental results, simulation results which shows that the proposed algorithm has the reliability and validity.