[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]
As computers become popular,users frequently use digital images. Every image does not fit to a user's requested size. Users might magnify or reduce an image to satisfy their request. This thesis deals with the problem that what kind of method to magnify an image can make a higher quality image than a image magnified with other method. We propose the high quality image magnification method based on an analysis of features of magnified images and the relationship between the features and image quality. Users can magnify an image with high quality, applying the method proposed in this thesis.
At first,we examine how image features vary when users magnify an image. The simplest way to magnify an image is to magnify pixels of it. We call it the simple magnification method in this thesis and call an image magnified with the simple magnification method a simple magnified image. One of the major problem of a simple magnified image is that it has jaggy edges. ``Jaggy'' means that border lines of regions in a simple magnified image are not smooth but jagged. We make two kinds of images to detect an appearance of jaggy edges. One kind of them has vertical resolution decreased and the other has horizontal resolution decreased. We can detect an appearance of jaggy edges in these images by analyzing the distributions of energy along concentric circles with the radius r and lines at the angle with the horiz ontal direction in Fourier power spectrum. Energy of a simple magnified image decrease in a high frequency part with a large value of r. An image in which vertical resolution decreased keeps original vertical edges and loses original slant and horizontal edges. It contains horizontal jaggy edges. Because vertical resolution and horizontal resolution usually decrease in a same ratio when users magnify an image, a simple magnified image loses slant edges most and contains vertical and horizontal jaggy edges.
Then,we investigate the relationship between the image features and image quality that affect an impression to human vision. We count the decrease of energy by decreased resolution and the increase of energy by jaggy edges in Fourier power spectrum. And we compare its result with the questionnaire survey about the quality of the images whose resolution decreasing vertically or horizontally. As a result,images regarded as poor quality lose much energy in a high frequency part and contain much energy by jaggy edges.
Therefore,there are two requirements to make a high quality image from a simple magnified image. The requirements are:
The program judges whether a square which is a simple magnified pixel of an original image makes jaggy edges or not,based on the color difference between the square and surrounding areas. If the square is regarded as a cause of jaggy edges, they are replaced with slant edges.
We process simple magnified images with our proposed method and analyze Fourier power spectrum about the processed images. According to the analysis,our proposed method satisfies the requirements to make a high quality image. In addition,we also analyze Fourier power spectrum about images processed with the averaging filter and with the median filter. The averaging filter does not satisfy the requirements at all and lowers the image quality. The median filter does not satisfy the requirement of recovering original slant edges.
At last in this report,we send out questionnaires about the quality of four kinds of images: simple magnified images,images processed with out proposed method, images processed with the averaging filter and images processed with the median filter. The result supports our proposed method and confirms its advantage.