Research On High Grayscale Image Generation And Exposure Fusion Technique

Much important information about details of images is contained in grayscale levels. And the lost details can be recovered by grayscale improvement technology, which produces clearer images with richer details. In recent decades, high grayscale image generation and visualization technique has become a focus in the area of image processing, including Grayscale Super-Resolution (GSR) and High Dynamic Range (HDR) two aspects. So far, research on GSR is rare and images obtained by the traditional methods still have less grayscale levels. The results of HDR image generation and display rely heavily on differently exposed images. To address these problems, we propose several new methods including GSR, HDR generation and exposure fusion methods.In the GSR method, we first propose to estimate a float number from all the integer measurements for the sum of this float number and a random variable with a given distribution. Then, a photographing apparatus with fine-tunable indirect fill lights is designed and realized, obtained from which the images/videos are used to increase grayscale precision. To validate this approach, we test on the simulated data and the real-world data. The experiment results show that the root mean square of errors between the reconstructed image and the reference image decreases with increasing the number of low grayscale images and finally converges. When applying into the real-world data, the results show that the reconstructed images have richer grayscale.Based on analyzing the relationship between the intensity value and the irradiance, we propose the linear model between the intensity values and the logarithm values of irradiance in the small neighborhood. The optimal model is designed with the intensity values of HDR image as the variable, which ensures that in the neighborhood the intensity change of HDR image be consistent with the maximum change in all Low Dynamic Range (LDR) images. The results show that the proposed method can generate an HDR image from fewer differently exposed images without estimating the camera response function. Besides, compared to the summation method, images obtained by the proposed method are closer to the logarithm values of irradiance.In the exposure fusion method, two improvements are proposed. First, weights are defined based on the background context to make the fused image smooth. Second, the enhancement method by removing the background context is proposed to compress the difference between different backgrounds and simultaneously increase the local contrast based on the linear model between the intensities before and after enhanced. To evaluate the results objectively, we propose the Standard Derivation of Gradient Magnitude (SDGM) as an objective index. The results show that in the images obtained by the proposed method, the local contrast in the dark and bright regions is higher and SDGM is smaller, indicating that the local contrast has been enhanced and the global appearance has been kept.We also propose the exposure fusion method based on fuzzy C-means clustering, in which all pixels are divided into a collection of two fuzzy clusters. Based on the degree to which each pixel belongs to the normal exposure cluster, a guided image is constructed and is combined with the Globally Optimized Linear Windowed Tone-Mapping method to obtain the fused image. The results show that the proposed method can increase the local contrast and the global brightness and that compared to6different tone mapping methods and1exposure fusion method, images obtained by the proposed method have bigger entropy and smaller SDGM.Methods in this paper have good performance in high grayscale image generation and exposure fusion. The proposed GSR method can considerably increase the grayscale precision. The proposed HDR generation method can produce an HDR image closer to the logarithm value of irradiance and its implementation is simple. Besides, exposure fusion methods depend less on input images, that is, fewer exposed images can also be used to produce good results.