KR20190070134A - Image Resizing apparatus for Large Displays and method thereof - Google Patents

Abstract

The present invention relates to an apparatus and a method for resizing an image of a large display, which can interpolate to intactly preserve details. In addition, the apparatus for resizing an image of a large display comprises: a parameter calculating unit for calculating an up-scaling parameter; an up-scaling unit for performing up-scaling by using the up-scaling parameter calculated in the parameter calculating unit; and an unsharp masking unit for performing unsharp masking on an image up-scaled in the up-scaling unit.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image resizing apparatus,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for adjusting an image size of a large display.

Signal scaling is an important and fundamental operation in image processing. Scaling is known as the scaling process [1-2].

This scaling process is widely used in multimedia tools such as medical imaging and digital photography.

The goal of interpolation is to estimate the data in a continuous signal model and then restore this function to the grid at the desired magnification [3].

Recent sizing methods have been applied to various fields. For example, image fitting of heterogeneous displays or screens can play an important role by skill rescaling [4-5].

In general, the scaling approach is categorized into two categories: the expansion process and the reduction process [7]. Interpolation is important for both processes.

There are several interpolation methods for image resizing. One of the best known and simple approaches is nearest neighbor interpolation.

Bilinear interpolation, bicubic interpolation and Lanczos interpolation are also widely used.

Although there are some interpolation methods, the most important role of interpolation is to restore and preserve the details of the restored image.

The existing interpolation method can not completely preserve the details, so a new interpolation method is suggested.

Registration No. 10-1795271 Publication No. 10-2013-0112500

[1] R. C. Gonzalez, R. C. Gonzalez and R. E. Woods, "Digital Image Processing", Addison-Wesley, Reading MA, (1992). [2] R. C. Tam and A. Fournier, "Image interpolation using unions of spheres," The Visual Computer, vol.14, (1998), pp. 401-414. [3] E. Maeland, "On the comparison of interpolation methods", IEEE Trans. Med. Image, vol. 7, no. 3, (1988), pp. 213-217. [4] J. A Parker, R. V Kenyon and D. E Troxel, "Comparison of interpolating methods for image resampling", IEEE Trans. Med. Image, vol. 2, no. 1, (1983), pp. 31-39. [5] R. Adipranata, E. Cherry, G. Ballangan and R. P. Ongkodjojo, "Fast Method for Multiple Human Face Segmentation in Color Image", International Journal of Advanced Science and Technology, (2009), pp. 19-32. [6] D. Bhattacharyya, A. Roy, P. Roy and T.-h. Kim, " Receiver Compatible Data Hiding in Color Image ", International Journal of Advanced Science and Technology, (2009), pp. 15-24. [7] B. Bhanu, J. Peng, T. Huang, and B. Draper, "Introduction to the Special Issue on Computer Vision and Pattern Recognition," IEEE Trans. Syst., Man, Cybern. B, Cybern., Vol. 35, no. 3, (2005), pp. 391-396. [8] J. Wang, G. Jeon, and J. Jeong, "Efficient adaptive de-interlacing algorithm with awareness of closeness and similarity", SPIE Optical Engineering, vol. 51, no. 1, (2012), pp. 017003. [9] X. Chen, G. Jeon, and J. Jeong, "A filter switching interpolation method for deinterlacing," SPIE Optical Engineering, vol. 51, no. 10, (2012), pp. 107402. [10] S. J. Park, G. Jeon, and J. Jeong, "Covariance-based adaptive deinterlacing method using edge map," in Proc. IEEE IPTA2010, Paris, France, (2010), pp. 166-171. [11] S. J. Park, G. Jeon, and J. Jeong, "Deinterlacing algorithm using edge direction from analysis of DCT coefficient distribution", IEEE Trans. Consumer Electronics, vol. 55, no. 3, (2009), pp. 1674-1681. [12] G. Jeon, SJ Park, Y. Fang, M. Anisetti, V. Bellandi, E. Damiani, and J. Jeong, "SPIE Optical Engineering, vol. 48, no. 12, (2009), pp. 127005. [13] J. Wu, Z. Xu, G. Jeon, X. Zhang, and L. Jiao, "Arithmetic coding for adaptive weight-context classification", Signal Processing: Image Communication, vol. 28, no. 7, (2013), pp.727-735. [14] X. Chen, G. Jeon, and J. Jeong, "Voting-based directional interpolation method and its application to still color image demosaicking", IEEE Trans. Circuits and Systems for Video Technology, vol. 24, no. 2, (2014), pp. 255-262. [15] X. Chen, G. Jeon, J. Jeong, and L. He, "Multidirectional weighted interpolation and refinement method for Bayer pattern CFA demosaicking", IEEE Trans. Circuits and Systems for Video Technology, vol. 25, no. 8, (2015), pp. 1271-1282.

SUMMARY OF THE INVENTION The present invention has been made to overcome the above problems, and it is an object of the present invention to provide an apparatus and method for adjusting an image size of a large display that can be interpolated to fully preserve details.

The apparatus of the present invention as described above includes a parameter calculation unit for calculating an up-scaling parameter; An up-scaling unit that performs up-scaling using up-scaling parameters calculated by the parameter calculator; And an unsharp masking unit for performing unsharp masking on the upscaled image in the upscaling unit.

Further, the method of the present invention includes the steps of: (A) calculating a up-scaling parameter by a parameter calculating section; (B) performing an up-scaling by using an up-scaling parameter calculated by the up-scaling unit; And (C) performing an unsharp masking on the upscaled image in the upscaling unit.

According to the present invention as described above, interpolation is enabled so as to preserve the details completely.

FIG. 1 is a flowchart illustrating a method of adjusting an image size of a large display according to an exemplary embodiment of the present invention.
Figure 2 compares the original blurry image with the corresponding UM process image.
3 is a block diagram of an apparatus for controlling the size of a large-sized display according to an embodiment of the present invention.
Figure 4 shows 20 LC images.
Figures 5 to 7 are restored images.
Figures 8 and 9 show the SSIM results for # 1 and # 3 images for comparison.
Figure 10 shows a comparison of the visual performance of the enlarged area.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings.

First, the terminology used in the present application is used only to describe a specific embodiment, and is not intended to limit the present invention, and the singular expressions may include plural expressions unless the context clearly indicates otherwise. Also, in this application, the terms "comprise", "having", and the like are intended to specify that there are stated features, integers, steps, operations, elements, parts or combinations thereof, But do not preclude the presence or addition of features, numbers, steps, operations, components, parts, or combinations thereof.

In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Image resizing is one of the key techniques for managing low-resolution images as high-resolution images.

FIG. 1 is a flowchart illustrating a method of adjusting an image size of a large display according to an exemplary embodiment of the present invention.

A low-resolution original image A _{i , j} having a width and a width twice as large as a high-resolution original image B _{i , j} having a height H and a width W can be obtained. Where i is the number of rows and n is a natural number, and j is the number of columns.

At this time, assuming that a high resolution original image B _{i , j} is obtained directly in a size of 2 × 2W, B _{2i, 2j} = A _{i , j} .

Where B _{i , j} is a high-resolution image. Now, we can interpolate the interlace grid B _{2i + 1} , _{2j + 1} from the grid B _{2i, 2j} = A _{i , j} .

To this end, the parameter calculator calculates the up-scaling parameter p (S100). The up-scaling unit performs image up-scaling using the up-scaling parameter as shown in Equation (1) (S200).

(1)

Where i is the number of rows, j is the number of columns, and k and l are the lattice parameters.

Assuming that a given image can be regarded as a locally fixed, independent, and uniformly distributed Gaussian model, the upscaling parameter p may be computed as follows by the standard least squares method. In the following Equation 2, the arrow shown above p is a vector representation of p.

(2)

는 고해상도 이미지 B_{i ,j}의 국소 공분산이다. C^-1은 C의 역행렬이다. Here C = [C _kl ] and

Is the local covariance of the high resolution image B _{i , j} . C ^-1 is the inverse of C.

In Equations (1) and (2), image upscaling is performed.

This technology can be widely used, such as image deinterlacing, noise reduction and demosaicing, to increase the size of grayscale or multi-channel images.

After the image up-scaling is performed, the unsharp masking unit performs unsharp masking (UM) (S300).

The unsharp masking process is an image enhancement process. The upscaling process may include interpolation and estimation. However, defects can occur because the original image size is half of the resulting image.

In addition, the upscaling image is generally blurred as the edge detail is lost. Thus, the unsharp masking process improves images that are less blurry than the upscaled image. The unsharp masking process is performed by the following equation (3).

(3)

Where k is an amount parameter that controls the unsharp masking effect.

If k is larger, more unsharp masking effects are applied. lpf is a low pass filter, and lpf (B) is a result image after low pass filtering.

는 언 샤프 마스킹된 결과 영상이다.B is the upscaled input image,

Is an unsharp masked result image.

The lpf filter is expressed by Equation (4). Parameter B is the result of applying the UM effect.

The function imfilter transforms the two-dimensional image x into a filter [1 1 1; 11 1; 1 1 1] / (h + 8).

(4)

Here, the weighting parameters k and h are set to 0.5 and 2, respectively. Figure 2 compares the original blurry image with the corresponding UM process image.

3 is a block diagram of an apparatus for controlling the size of a large-sized display according to an embodiment of the present invention.

Referring to FIG. 3, the apparatus for adjusting the size of a large display according to an exemplary embodiment of the present invention includes a parameter calculator 100, an upscaling unit 200, and an unsharp masking unit 300.

A low-resolution original image A _{i , j} having a width and a width twice as large as a high-resolution original image B _{i , j} having a height H and a width W can be obtained. Where i is the number of rows and n is a natural number, and j is the number of columns.

At this time, assuming that a high resolution original image B _{i , j} is obtained directly in a size of 2 × 2W, B _{2i, 2j} = A _{i , j} .

Where B _{i , j} is a high-resolution image. Now, we can interpolate the interlace grid B _{2i + 1} , _{2j + 1} from the grid B _{2i, 2j} = A _{i , j} .

For this, the parameter calculation unit 100 calculates the up-scaling parameter p. The up-scaling unit 200 performs image up-scaling using the up-scaling parameters as shown in Equation (1).

Assuming that a given image can be regarded as a locally fixed, independent, and uniformly distributed Gaussian model, the upscaling parameter p is computed by the standard least squares method using equation (2).

After the image upscaling is performed, the unsharp masking unit 300 performs unsharp masking (UM).

The unsharp masking process is an image enhancement process. The upscaling process may include interpolation and estimation. However, defects can occur because the original image size is half of the resulting image.

In addition, the upscaling image is generally blurred as the edge detail is lost. Thus, the unsharp masking process improves images that are less blurry than the upscaled image. The unsharp masking process is performed in Equation (3).

We tested the experiment below to demonstrate the quality of the approach presented.

Use MATLAB software with Intel (R) Core (TM) i5 CPU M460 @ 2.53GHZ processor. We used 20 LC data sets between # 1 and # 20 images.

Figure 4 shows 20 LC images. The size of the test image is 720x540.

Three benchmark methods were used to compare visual performance: nearest NN (NN), Bilinear (BI), and Bicubic (BC) methods.

We used three quality assessment tools: mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) metrics. To assess the performance of objective quality and visual performance, three indicators were used.

The PSNR is calculated as follows.

(5)

Where MSE represents the mean square error and MAX represents the maximum possible intensity of a given image. When a pixel is represented using n bits per sample, MAX is 2n-1. Generally, the acceptable PSNR level is 30 to 50 dB for an 8-bit image.

(6)

In the case of a 16-bit image, 60 to 80 dB is preferable. MSE is the mean squared error, where ori and res represent the original image and the resulting image.

The SSIM index is designed to assess the quality between the original image and the reconstructed image. The SSIM is used to estimate the similarity between two images.

Since the original image is used for evaluation, the SSIM is a complete reference metric. SSIM is intended to enhance existing quality assessment tools such as PSNR and MSE.

The restored image is displayed in 5-7. In all figures, image (a) represents an original cropped image of 256x256 size. In the same way, images (b), (c), (d) and (e) are the results of the nearest neighbors method, the bilinear method, the bicubic method and the proposed method. All figures show that the proposed method generates subjective interesting results when compared to the benchmark method.

For example, stair artifacts are displayed in the nearest neighbors result, blur effects are shown in bilinear and dipole, and the proposed method preserves the details well. This can be seen in the window and door areas of FIG.

Figure 6 shows the "bikemen" image. The resulting image by the proposed method restores the characters of the jersey, helmet and roller skates.

Figure 7 shows a roof image. Figures 6 (b), 6 (c) and 6 (d) do not preserve the details of the roof, and the proposed method shows details.

Figures 8 and 9 show the SSIM results for # 1 and # 3 images for comparison. Four images (a) to (d) are the closest neighbors, bilinear, bicubic, and the result of the proposed method.

A result closer to "1" in the SSIM metric indicates a better image, and a result closer to "0" indicates a worse image. As you can see, the proposed method provides the best SSIM results across a given image. In Fig. 9, the SSIM value in the edge near the mountain region is high. This means that the proposed method has better performance than the other methods.

Objective performance comparisons are provided in Table 1-3 using PSNR, MSE, and SSIM metrics.

(Table 1)

(Table 2)

(Table 3)

In terms of the PSNR metric, the proposed method outperforms other methods with quantities of 0.776, 0.375 and -0.537 dB. PSNR measurements provide better performance than the proposed method, but visual comparisons show that the proposed method yields better results. Similarly, MSE and SSIM metrics were used for performance evaluation.

The proposed method outperforms other methods with quantities of -19.392, -8.506 and 18.756 for the MSE metric, and 0.010, 0.024 and -0.011 for the SSIM metric.

Figure 10 shows a comparison of the visual performance of the enlarged area.

As mentioned earlier, the closest approach is one of the simplest ways to replace all pixels with a number of pixels of equal intensity.

Thus, the reconstructed image represents an unwanted zigzag, but all pixels are original. Recent stochastic artifacts are clearly displayed in the results of the lean interpolation method. The bilinear method and the bicubic method provide blurred images. However, the proposed method generates a pleasant result image that preserves the details. Both methods produce a continuous change in results, even wherever they are.

The original pixel has a discontinuous change. Both methods therefore reduce undesirable contrast in the linear pattern image. The bicubic method is known to be superior to the bilinear method, but the difference is negligible and the complexity is higher.

In the present invention, an image size adjustment method is proposed.

The purpose of the proposed method is to resize the image for a given display device. The proposed method consists of three steps: S1: self-learning based on least squares method, S2: image upscaling, and S3: image unsharp masking. The parameter p is calculated in S1 and the parameters k and h are calculated in S3. Experimental results show that the proposed approach provides satisfactory results in terms of PSNR, MSE and SSIM metrics. Subjective performance comparisons also indicate that the proposed method is superior to the other methods.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments of the present invention are not intended to limit the scope of the present invention but to limit the scope of the present invention. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents thereof should be construed as being included in the scope of the present invention.

100: parameter calculating unit 200: upscaling unit
300: Unsharp masking portion

Claims (15)

A parameter calculation unit for calculating an up-scaling parameter;
An up-scaling unit that performs up-scaling using up-scaling parameters calculated by the parameter calculator; And
And an unsharp masking unit for performing unsharp masking on the upscaled image in the upscaling unit. The method according to claim 1,
Wherein the parameter calculator calculates an up-scaling parameter using a standard least-squares method. The method according to claim 1,
The parameter calculator calculates the upper-original image A _{i , j} obtained by obtaining a low-resolution original image A _{i , j} in which the width and width are doubled in the high-resolution original image B _{i , j} having the sizes of the height H and the width W _, Where i is the number of rows and j is the number of columns, each of which is a natural number. The method according to claim 3,
Wherein the up-scaling unit performs up-scaling using the following equation (1).
(1)

Where i is the number of rows, j is the number of columns, each is a natural number, and k and l are lattice variables. The method of claim 4,
Wherein the parameter calculator calculates the upscaling parameter p using Equation (2) below.
(2)

Where the arrow above p is the vector representation of the upscaling parameter, C = [C _kl ] and

Is the local covariance of the high-resolution image B _{i , j} , and C ^-1 is the inverse of C. The method according to claim 1,
Wherein the unsharp masking unit performs unsharp masking using Equation (3).
(3)

Where k is a control weighting parameter that controls the unsharp masking effect.
lpf is a low pass filter, lpf (B) is a result image after low pass filtering, B is an upscaled input image,

Is an unsharp masked result image. The method of claim 6,
The low pass filter lpf is expressed by the following equation (4).
(4)

Here, x represents an input image, and h represents a weight parameter. The method of claim 7,
Wherein the control amount weight parameter k is 0.5 and the weight parameter h is 2. (A) calculating a parameter up-scaling parameter;
(B) performing an up-scaling by using an up-scaling parameter calculated by the up-scaling unit; And
(C) performing an unsharp masking on an upscaled image in the upscaling unit by an unsharp masking unit. The method of claim 9,
Wherein the parameter calculating unit calculates an upscaling parameter using a standard least squares method in the step (A). The method of claim 9,
In the step (A), the parameter calculation unit may calculate a low resolution original image A _{i , j} having a width and a width that is twice as high as a high resolution original image B _{i , j} having a height H and a width W, Calculating an upscaling parameter using a local covariance, where i is the number of rows and j is the number of columns, and each is a natural number. The method of claim 9,
Wherein the unsharp masking unit performs unsharp masking using Equation (3) in the step (C).
(3)

Where k is a control weighting parameter that controls the unsharp masking effect.
lpf is a low pass filter, lpf (B) is a result image after low pass filtering, B is an upscaled input image,

Is an unsharp masked result image. The method of claim 12,
Wherein the low pass filter lpf is expressed by the following equation (4).
(4)

Here, x represents an input image, and h represents a weight parameter. The method of claim 13,
Wherein the control amount weight parameter k is 0.5, and the weight parameter h is 2. (A) calculating a parameter up-scaling parameter;
(B) performing an up-scaling by using an up-scaling parameter calculated by the up-scaling unit; And
(C) Unsharp masking unit performs unsharp masking on the up-scaled image in the up-scaling unit. A computer-readable recording medium having recorded thereon a method of adjusting an image size of a large-sized display.

Images