Research On Near-infrared Image Enhancement Algorithm

In recent years,with the growing social needs of infrared image and the improving requirement for the quality of infrared image,the research of infrared image becomes a hot topic among the scientific community.The formation of infrared image depends on infrared radiation of target objects,but the radiation intensity is weakened when transmitting in various media,causing the quality of original infrared image is poor.Because of this,the original infrared image can't meet the requirement in the field of military,medical and civilian.However,the technique of hardware equipments is not good enough to radically solve the above problem.Therefore,using software algorithm to enhance the infrared image is necessary.In the discussion of this paper,some problems,like the existence of blind pixels and the nonuniformity noise,low image contrast and so on,are analyzed according to the principle of Infrared focal plane array imaging.Then the original infrared image is enhanced from two stages,image preprocessing and image enhancement.In the process of image preprocessing,the blind pixels are detected by the combination of full window detection method and sliding window detection method and compensated using the spatial and temporal correlation compensation method.And then the stripe noise,caused by the nonuniformity of the detector,is eliminated using the two-point correction method.The results of the experiment indicate that more than 99%of the blind pixels in image can be detected and compensated;and the stripe noise can well be eliminated.In the process of image enhancement,the principle of traditional image enhancement algorithms are analyzed and the defects of them existing in infrared image processing are summed up.Then combined with the practical problems encountered in this research project,a kind of near infrared image enhancement algorithm based on steerable pyramid is put forward(infrared image can be divided into three kinds,near infrared,middle infrared and far infrared,according to the wavelength of the infrared.near infrared image is one of them).First,the preprocessed near infrared image is decomposed into multi-scales using steerable pyramid model,then the low-frequency coefficients and high-frequency coefficients of the image are obtained.Multi-scales of the image can better retain the contour information of the image,and steerable pyramid supports polydirectional high-pass filtering and rotation invariance,so the detail of the image can be better retained.In order to improve the contrast of the image,the low-frequency coefficients are transformed nonlinearly using fuzzy-set theory.The high-frequency coefficients are dealt with threshold method to separate the detail information of the image and the noise information.Next,through inverse transformation,the image is reconstructed.At last,unsharp masking is used to highlight the details of the image.The processed image can not only provide better visual effect,but the contrast gain can be kept above 6.20,up to 7.12.And the information entropy is increased above 1.20 times,up to 1.67 times.In the process of image reconstruction,image sampling and interpolation methods are used.The traditional zero interpolation and bilinear interpolation would lead to losing image edges,causing serious image distortion.So an adaptive interpolation algorithm is put forward.During the process of down-sampling,the ratio between removed pixels and remaining pixels is recorded.Then during the process of interpolation,the recorded ratio is used to maintain the relationship between the new insertion pixels and remaining pixels.So the lose of image detail is decreased.