Research Of Image Enhancement Based On Probability Theory

The application of digital images has rapidly increased with growing public con-sumption of entertainment and communication appliances, such as digital TV s, digital cameras, scanners, mobile phone cameras, and personal media players. The expecta-tion of a higher image quality prompts researchers to develop cutting-edge techniques for image enhancement. Histogram equalization has been one of the most widely used techniques due to its effectiveness and simplicity in contrast enhancement.The conventional HE algorithm has several drawbacks. First, it often causes un-pleasant visual artifacts. When a histogram bin has a very large value, the transfor-mation function gets an extreme slope. This can cause contrast overstretching, mood alteration, or contour artifacts in the output image. Second, especially for dark im-ages, HE transforms very low intensities to brighter intensities, which may boost noise components as well, degrading the resulting image quality. Third, the level of contrast enhancement cannot be controlled, since the conventional HE is a fully automatic algo-rithm without any parameter. It is also especially difficult to achieve a well-balanced enhancement effect over different parts of an image, e.g., the background and detail parts of the image.To solve the problems above, this paper is a study on the theory of the following five aspects:1) a piecewise histogram equalization based image enhancement;2) an entropy preserving histogram specification with adaptive brightness;3) an equalized closest-mode filter based on smoothed local histogram;4) an equalized median filter and dominant-mode filter based on smoothed local histogram; The paper from chapter3to chapter5are corresponding to these five research topics.Main contributions are listed below.1. A new algorithm (Piecewise HE) is proposed based on histogram equalization, which combines the benefits of contrast stretching and existing histogram equalization. First, it is strictly proved that the entropy of image must decrease when the swallow of bin in the histogram exist. By analyzing the relation of swallow and entropy, a sufficient and necessary condition is obtained to prevent the entropy reduction. The swallow of saturation and brightness is avoided in the enhanced image by a piecewise linear mapping function, which combines with the mean and standard deviation of the histogram, and the parameters selection of Piecewise HE is explored when over-enhancement existed. The experimental results have demonstrated that the proposed method can enhance the images effectively, which avoids the swallow and over-enhancement problems of the existing histogram equalization methods. The proposed method not only preserves the original details, but also has a more natural visual effect in the enhanced image.2. A new method is proposed about histogram specification to overcome those drawbacks of HE. Our method (EPHEMP) makes the histogram as flat as possible by the fundamental idea of HE, and by the variation algorithm, finds a suitable target histogram maximize the mean brightness under the constraint that the entropy is constant, at last transforms the original histogram to that target one by histogram specification. Comparing to the existing methods including HE, BBHE, DSIHE, MMBEBHE, and BPHEME, experimental results show that EPHEMP can not only preserve the entropy, but also enhance the contrast of the image effectively. And hence it is possible to be used for many commercial purposes such as consumer and professional electronic products.3. An equalized closest-mode filter method(CMHEF) is proposed based on smoothed local histogram. Since an important limitation is that histogram-based filters often in-crease the sharpness of the edges they preserve, the over-sharpening edge leads to halo, gradient reversals or other artifacts in the output. And all histogram-based computa-tions face a computational challenge when dealing with large neighborhoods. After we formulate the closest-mode filter, local histogram equalization is introduced. By the derivative look-up table and map table, it is possible to compute the equalized closest-mode filter without the iterative procedure. Comparing to the closest-mode filter, our methods can preserve the edge, details and textures of image, in addition enhance the contrast of different parts of an image.4. An equalized median filter method(MHEF) is proposed based on smoothed local histogram. For dark images, traditional HE transforms very low intensities to brighter intensities, which may boost noise components as well. Despite median filter is best known for its salt and pepper noise removal aptitude. Unfortunately, none of median algorithms is isotropic, and none of them provides weights that drop off smoothly from the center of the neighborhood. So we formulate the median-mode filter, and local histogram equalization is introduced. By a new model of combining the integral look-up table and map table, our method of computing the equalized median takes constant time per output sample, independent of the neighborhood size. Comparing to the median filter, our methods can preserve the edge smoothing and remove noise for dark image, which better enhance the contrast of an image. In principle, a more accurate result is possible by using a robust criterion to choose among the local modes. Based on the above two models, we select the mode corresponding to the largest population of samples in the equalized smoothed local histogram. Similar to the methods above, local histogram equalization is introduced to smoothed local histogram. By the different procedure to find look-up tables, the most mode in the equalized smoothed local histogram is obtained without the iterative procedure. The experiment results show our methods can not only remove the noises very effectively and sharpen the edge, but enhance the contrast of different smooth regions of an image. And hence the proposed method has better robustness and more widely applied range.