| Popis: |
In this research we did quantification of contrast level for various general and biomedical images. Here we considered two novel techniques under evaluation first one is Histogram Flatness Measure (HFM) and second one is Histogram Spread (HS). In case of HFM the values of the measure are found to be inconsistent in the sense that even for low contrast images also the value of HFM are higher than original images being inconsistent to identify whether it is high contrast or low contrast image with respect to original image. When it comes to HS the values of the measure are coming as for high contrast images high values of HS and low contrast images low values of HS with respected to the value of Original Image. Here we observed the values in terms of taking the images of high contrast at certain level to fully high contrast (histogram equalized images). For all the high contrast images HS found to be high values and low contrast images HS found to be low values and being highly consistent and specifically useful than HFM. The standardization of HS can be useful in database management, visualization, image classification. As par the images we put into application for the evaluation of Novel metric ‘HS’, then we may standardize the notion “High contrast high value and low contrast low value”. Then we did want to analyze the images of high contrast which follow the notion along with images of histogram equalized high contrast images for every image under consideration, in terms of image contrast enhancement Image Quality Assessment (IQA) measures, and we would like to find out the image whether if it is higher level of high contrast or lower level of high contrast with respected to the fully high contrast (histogram equalized) image, so that we can have an idea of the given partially high contrast (towards histogram equalization) image that how much near it is to Fully high contrast image. One advantage of this observation is that, if the image is high contrast and if it is far from fully high contrast image, then we can become cautious especially In the case of Biomedical Images, Cosmological images that the images have to be carefully preserved. Here we have found the Computational time in milliseconds for each IQA measure for all the images so that we can have an idea of which measure we can choose for any hardware design of the performance measure by using tradeoffs basing on the required application of interest of any high speed Digital Image Processor Hardware Implementation for any Real time Medical Device. The image quality measures we considered are Peak Signal–to-Noise-Ratio (PSNR), Mean Absolute Error (MAE), Absolute Mean Brightness Error (AMBE), Signal to Noise Ratio (SNR), Contrast to Noise Ratio (CNR), Universal Quality Index (UQI), Noise Quality Measure (NQM), Structural SIMilarity (SSIM), Mean SSIM(MSSIM), Information Fidelity Criterion(IFC), Visual Information Fidelity (VIF), Visual Information Fidelity in Pixel Domain (VIFP), Visual Signal-to-Noise Ratio (VSNR), Wavelets Based SNR (WSNR), Feature similarity Metric (FSIM), Riesz Transform FSIM (RFSIM). One more observation we made is that partially high contrast images are having low computational time of performance metrics when compared to fully high contrast (histogram equalized) images. So in some applications, where no need of fully high contrast images, we can utilize these reasonably high contrast images instead of fully high contrast images for reducing the computational time and also it is highly useful in the case of Digital Image Processing hardware design for saving time and to get a high speed processor, which is highly useful observation. All the research is done in MATLAB 8.3 R2014a programming. |
