Computer graphics processing and selective visual display system – Computer graphics processing – Attributes
Reexamination Certificate
2001-02-23
2004-11-09
Bella, Matthew C. (Department: 2672)
Computer graphics processing and selective visual display system
Computer graphics processing
Attributes
C345S426000, C345S215000, C345S215000, C345S215000, C345S215000, C345S600000, C345S660000, C345S663000
Reexamination Certificate
active
06816166
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method and an apparatus for converting the resolution of an image consisting of a set of pixels that are represented by digital values.
2. Description of the Related Art
So far, it has been assumed that such an image that is taken from real world to a photo and digitized is the object of resolution conversion, and that a change of pixel values is continuous and a discontinuous boundary has a sufficient size. Most conventional image expansion techniques aim at natural images, which rarely have stepped edges such as thin lines in themselves. This is mainly because information is recorded, while the stepped edge is transformed, by the lowpass effects of an image pickup apparatus, into a form represented by a sigmoid function (a differentiable, continuous function represented by f(x)=1/(1+(e
−x
)). Therefore, so far, improvements have been made to find how to make the outline, which would become blurrier than necessary, appear natural, with the assumption that the original image was obtained by sampling a lowpass filtered image.
For resolution conversion which requires the least computations and can be easily implemented, the replica and nearest neighbor methods are pointed to among conventional ones. The replica method is the simplest one by which an image can be expanded (n+1)
times by copying the same pixels every n-th pixel. On the other hand, the nearest neighbor method is the one by which an image is expanded by copying the pixel of the original image that is closest to the coordinates obtained after the resolution conversion. Both the replica method and the nearest neighbor method provide substantially the same effects; i.e. the mixing of colors among pixels does not occur, and color tones are completely held.
Also, two methods, i.e. bilinear interpolation and bicubic (cubic convolution) interpolation are well known for the conversion of resolutions. With bilinear interpolation, the coordinates of a pixel point for a resultant image are inversely mapped to the coordinates of the original image. Then, the adjacent pixels of the original image (four surrounding points, two points on both sides, or only one point located at the identical coordinate) are weighted, by using distances, to obtain an average value, which is subsequently regarded as the color of the resultant image. On the other hand, bicubic interpolation is extended up to 2 rounds (16 points) in the vicinity based on the same idea as bilinear interpolation. According to this method, it is assumed that the one-order differential continuity and the change of values (i.e. slope) of the original image are sufficiently moderate. Colors are enhanced by the weights of parameters, which provides the advantage of making them clearer than bilinear method.
In addition, a method that generalizes the above interpolation methods and that requires much more computations, is identified as multirate system. In general, a digital signal processor (DSP) is required in order to implement this multirate system. The basic framework of the multirate system is to apply lowpass filtering after up-sampling with zero-value interpolation, and further to apply down-sampling with decimation to obtain a predetermined expansion factor. This framework can theoretically comprehend the bilinear interpolation and the bicubic interpolation by using the frequency characteristics of the lowpass filter. Practically, implementations called polyphase configuration or filter bank configuration are often used in order to reduce the computational load.
However, according to the replica method and the nearest neighbor method, the pixel of an original image is simply held at a new spatial sampling point, and the expanded line width differs depending on the coordinate positions. Since the sensitivity to frequency components of human eyes is high to low frequencies for the angle unit, a serious problem occurs in the readability of a line width in these methods.
Further, according to the bilinear interpolation method, when the coordinates that are determined for an inversely projected image are located at the same distance on either side of a one-dot width line, the resultant image will invariably represent a line as wide as two lines of a half a color. Therefore, problems occur relative to the uniformity of colors, the legibility of characters and the reproduction fidelity of colors. Although an image in a photo may appear to be satisfactory, the overall impression is of a blur.
Furthermore, the bicubic interpolation method has the same problem as does bilinear interpolation method, where one line is changed into two lines half a color. Thus, the accurate reproductivity of colors of the screen image of a personal computer (hereinafter referred to as a PC) is still problematic. Also, there is a slight occurrence of ringing at sharp boundaries between middle tone colors.
Moreover, when any given lowpass filter is employed for the above multirate system, checkerboard distortion may appear in a resultant image, thereby imposing constraints upon filter designs. That is, constraints are placed on filter designs depending upon the images that are to be processed and required filter conditions, such as, the filter characteristics for a passing band or blocking band must be flat, or a filter for down-sampling be separately provided in order to introduce an average operation. It is preferable that an FIR (Finite Impulse Response) filter be used to maintain the linear phase of an image, however, typically a higher order filter is required to obtain an image quality that is higher than what is acquired by the bilinear or bicubic interpolation method. So it is believed necessary to use a dedicated DSP. In addition, as a problem associated with using the higher order filter, what has a thin line structure as a font, is expanded more than necessary to appear.
In U.S. Pat. No. 5,594,676, a method for scaling is proposed wherein the filtering ranging from 3×3 to maximum 65×65 is performed by changing the number of pixels used in the interpolation according to the position of a pixel. However, on a PC screen subject to the present invention, since there are a number of changes of a stepped edge with one pixel width, the above prior methods can not perform a proper resolution conversion. Now let's consider the reason in terms of a multirate system.
FIG. 13
depicts a multirate system performing U/D-fold sampling-rate conversion. First, converting an input x[n] comprising n sample points to n·U data string using an up-sampling( ↑U)
201
, then passing through a lowpass filter
202
, and finally obtaining m=n·U/D resultant data string y[m] using down-sampling(↓D)
203
. In up-sampling, U−1 zero values are added per one of x[n] data.
FIG. 14
depicts an example of one-dimensional signals using U=3 for up-sampling. An upper left diagram shows the input string, while an upper right diagram shows the first twenty of data string that were increased three times with the interpolation of zero values. A horizontal axis of these diagrams represents a discrete sampling number, while a vertical axis represents an amplitude (level). A lower left diagram depicts a result of x[n] after FFT (Fast Fourier Transform), while a lower right diagram depicts a result of up-sampled 3·n data string after FFT. A horizontal axis of these diagrams represents a normalized frequency [Hz], while a vertical axis represents a spectral amplitude represented by square of an absolute. Both of them normalize the frequency to [0,1], wherein 0 corresponds to a lower frequency. As can be seen in the lower right diagram, the frequency components of an original image is reduced to one U-th., while at the same time imaging components appear in the higher frequency domain.
FIG. 15
depicts an example of one-dimensional signals using D=3 for down-sampling. A horizontal axis and a vertical axis are the
Sekiya Kazuo
Shimizu Junya
Augspurger Lynn L.
Bella Matthew C.
Kinnaman, Jr. William A.
Tran Tam
LandOfFree
Image conversion method, image processing apparatus, and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image conversion method, image processing apparatus, and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image conversion method, image processing apparatus, and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3292900