KR20010045909A - filter coefficient generatig method for image scaling - Google Patents

Abstract

PURPOSE: A method for generating a filter coefficient for an image scaling is provided to quickly increase filter coefficients and to reduce the load of a hardware by storing filter coefficients for the interpolation points in a filter coefficient memory and accessing it during the scaling. CONSTITUTION: A scale function is sampled and stored in a coefficient memory. A window function is sampled and stored in a window memory to restrain the scale function. A filter coefficient for each interpolation point is read out from the coefficient memory referring to the scale factor, M. A window coefficient is read out from the window memory referring to the scale factor, M. A filter coefficient obtained by multiplying the scale factor, M with the window coefficient is stored in a filter coefficient memory. For each interpolation point, a filter coefficient stored in the filter coefficient memory is read out.

Description

Filter coefficient generatig method for image scaling

The present invention relates to an image scaler, and more particularly, to a method of generating filter coefficients capable of providing filter coefficients in a short time depending on the position of an interpolation point.

In accordance with the trend toward larger and more digitized display devices, display devices capable of supporting various resolutions have been commercialized. In such a display device, a resolution converter, that is, an image scaler, is required to adjust the output resolution.

Since the scale function used in the high quality image scaler is expressed as a polynomial, many operations are required. In particular, the filter coefficients multiplied by neighboring pixels are different depending on the interpolation point.

The hardware for calculating filter coefficients is simple when the scale function is expressed by simple equations such as bilinear, etc., but the scale function represented by the polynomial is designed considering the processing speed and size of the hardware for calculating the filter coefficients. Is needed.

An object of the present invention is to provide a filter coefficient generating method capable of providing filter coefficients in a short time depending on the position of an interpolation point in an image scaler.

1 is a graph showing a spline function, which is a kind of scale function.

FIG. 2 is shown to conceptually show sampling the scale function shown in FIG. 1.

3 is a conceptual illustration of a method of reading filter coefficients at any interpolation point w located in a section between 0 and 1. FIG.

4 schematically shows the contents of the filter coefficient memory.

FIG. 5 is a diagram for schematically showing the effect of the window function in the case of scale down.

6 is a block diagram showing the configuration of an apparatus for performing a filter coefficient generation method according to the present invention.

7 is a flowchart illustrating a scale up / down preparation process in the apparatus shown in FIG. 6.

FIG. 8 is a flowchart illustrating a filter coefficient output process in the apparatus of FIG. 6.

Filter coefficient generation method according to the present invention to achieve the above object

A filter coefficient generation method for image scaling for converting and outputting a resolution of an input image according to a predetermined scale factor M,

Sampling a scale function for scaling and storing the scale function in a coefficient memory;

Sampling a window function for limiting the scale function and storing the sampled window function in a window memory;

Reading filter coefficients for each interpolation point from the coefficient memory with reference to a scale factor M;

Reading window coefficients from a window memory with reference to the scale factor M;

Storing filter coefficients obtained by multiplying the filter coefficients by a window coefficient in a filter coefficient memory; And

And reading out and outputting filter coefficients stored in the filter coefficient memory for each interpolation point.

Hereinafter, the configuration and operation of the present invention will be described in detail with reference to the accompanying drawings.

1 is a graph showing a spline function, which is a kind of scale function. Basically, the width of the scale function is determined according to the number of pixels involved in interpolation, and as shown in FIG. 1, eight pixels are involved in interpolation. In addition, pixels that participate in interpolation are located at the zero crossing points on the horizontal axis.

Referring to FIG. 1, when the function value of the scale function is f (x) and the value of the pixel involved in interpolation is Q (x), an arbitrary interpolation point x between integers x and x + 1 The pixel value Q (x + w) at + w is expressed as follows.

Q (x + w) = f (n + w) * Q (x + n) + ,,, + f (2 + w) Q (x + 2) + f (1 + w) Q (x + 1) + f (w) Q (x) + f (w-1) Q (x-1) + f (w-1) Q (x-2) + ,,, + f (wn) Q (xn)

Where n is an integer representing the number of pixels involved in interpolation or the number of taps of the filter, f (n + w), f (2 + w), f (1 + w), f (w ), f (w-1), f (w-2), ,, and f (wn) are coefficients (hereinafter referred to as filter coefficients) that are multiplied by each of the pixels involved in interpolation.

The filter coefficient calculation process according to the present invention is performed as follows.

1) First, sample and save the scale function.

The scale function is sampled at equal intervals (power of 2) to ensure the subjective and objective image quality and stored in the coefficient memory.

When the scale function is symmetric with respect to the origin as shown in FIG. 1, only the left or right side is sampled from the origin to reduce the capacity of the scale function memory.

FIG. 2 is shown to conceptually show sampling the scale function shown in FIG. 1.

When the scale function is symmetric about the origin as shown in FIG. 1, only one side is sampled about the origin as shown in FIG. 2.

As shown in Fig. 2, a total of 128 (2 ⁷ ) small sections (hereinafter referred to as offsets) are divided, and the scale function value f (x) in each offset is sequentially counted in order from the origin. Stored in memory.

As shown in Equation 1, eight filter coefficients are required to obtain the pixel value Q (x + w) at an arbitrary interpolation point (x + w), and each filter coefficient is separated by a certain distance on the scale function. The separation distance of each filter coefficient depends on the scale factor M. In the case of outputting an image having a higher resolution than the input image, the separation distance of each filter coefficient is 1, and in the case of outputting an image having a lower resolution than the input image The separation distance of each filter coefficient is in inverse proportion to the scale factor M.

2) The filter coefficients at each of all interpolation points are read from the coefficient memory with reference to the scale factor M and stored in the filter coefficient memory.

The filter coefficients at each interpolation point according to the scale factor M are read by controlling the read address of the coefficient memory.

The interpolation point in scaling is located in one of the offsets shown in FIG. Accordingly, addresses in which filter coefficients corresponding to the interpolation point are stored in the coefficient memory may be automatically calculated through a predetermined equation when the address of the offset to which the interpolation point belongs is known.

3 is a conceptual illustration of a method of reading filter coefficients at any interpolation point w located in a section between 0 and 1. FIG.

The filter coefficients required to obtain the interpolation value at any interpolation point w located in the interval between 0 and 1 are f (3 + w), f (2 + w), f (1 + -w), f (w) , f (1-w), f (2- + w), f (3-w), f (4-w). They obtain the address of the offset to which the interpolation point belongs in the coefficient memory, and read the filter coefficients in order from the coefficient memory with reference to this. The read result is stored in the filter coefficient memory in the form of a bit stream. This is performed for all interpolation points, and the filter coefficients for each of all interpolation points are read and stored in the filter coefficient memory.

(1) In the case of scale up

First, the offset address to which the interpolation points belong is determined by referring to the scale factor M. When scaling up, the scale factor M is represented by a real number, and the interpolation point is at one of 32 offsets divided by 0 < When the address w of the offset to which each interpolation point belongs is determined, the following equation (2) determines an address (coeff address) to access the coefficient memory to read the filter coefficients for the corresponding interpolation point.

Address of the first filter coefficient coeff [0] coeff address1 = 3 * 32 + w

Address of the second filter coefficient coeff [1] coeff address2 = 2 * 32 + w

Address of the third filter coefficient coeff [2] coeff address3 = 1 * 32 + w

Address of the fourth filter coefficient coeff [3] coeff address4 = w

Address of the fifth filter coefficient coeff [4] coeff address5 = 1 * 32-w

Address of the sixth filter coefficient coeff [5] coeff address6 = 2 * 32-w

Address of the seventh filter coefficient coeff [6] coeff address7 = 3 * 32-w

Address of the eighth filter coefficient coeff [7] coeff address8 = 4 * 32-w

Here, the number 32 is the number of offsets divided in the interval 0 <x <1 of the scale function. As shown in FIG. 2, when the number of pixels involved in interpolation is H and the number of samplings is P, the number of offsets divided in each section of the scale function is represented by P / H.

Eight filter coefficients are read out from the coefficient memory for all interpolation points and stored in the filter coefficient memory in turn.

The contents stored in each address fc_address of the filter coefficient memory are as follows.

fc_address [0] = coeff_0 [0], coeff_0 [1], coeff_0 [2], coeff_0 [3], coeff_0 [4], coeff_0 [5], coeff_0 [6], coeff_0 [7]

fc_address [1] = coeff_1 [0], coeff_1 [1], coeff_1 [2], coeff_1 [3], coeff_1 [4], coeff_1 [5], coeff_1 [6], coeff_1 [7]

,

,

fc_address [m-11 = coeff_m-1 [0], coeff_m-1 [1], coeff_m-1 [2], coeff_m-1 [3], coeff_m-1 [4], coeff_m-1 [5], coeff_m- 1 [6], coeff_m-1 [7]

Here, m is an integer of 1 to 32, and coeff_m [0] means the first filter coefficient at the mth interpolation point.

4 schematically shows the contents of the filter coefficient memory. In the first address of the filter coefficient memory, the filter coefficients at the first interpolation point are stored in the form of a bit stream. Similarly, filter coefficients at corresponding interpolation points are stored according to the address order of the filter coefficient memory.

(2) In case of scale down

First, the position of the offset to which the interpolation points belong is determined by referring to the scale factor M. The scale factor M is represented by a real number and there are (1 / M-1) interpolation points between adjacent (1 / M) pixels.

When the address w of the offset to which each interpolation point belongs is determined, an address (coeff address) for reading a filter coefficient for the corresponding interpolation point is determined by Equation 3 below.

Address of the first filter coefficient coeff [0] coeff address1 = 3 * (1 / M) * 32 + w

Address of the second filter coefficient coeff [1] coeff address2 = 2 * (1 / M) * 32 + w

Address of the third filter coefficient coeff [2] coeff address3 = 1 * (1 / M) * 32 + w

Address of the fourth filter coefficient coeff [3] coeff address4 = w

Address of the fifth filter coefficient coeff [4] coeff address5 = 1 * (1 / M) * 32-w

Address of the sixth filter coefficient coeff [5] coeff address6 = 2 * (1 / M) * 32-w

Address of the seventh filter coefficient coeff [6] coeff address7 = 3 * (1 / M) * 32-w

Address of the eighth filter coefficient coeff [7] sf_address8 = 4 * (1 / M) * 32-w

Here, the number 32 is the number of offsets divided in each section of the scale function. As shown in FIG. 2, when the number of pixels involved in interpolation is H and the number of samplings is P, the number of offsets divided in each section of the scale function is represented by P / H.

By the way, the scale function in scale down substantially extends the scale function shown in FIG. 1 by 1 / M in the width direction. In addressing the coefficient memory, the interval of addresses for reading filter coefficients is 1 / M. Means reduced. Therefore, substantially in the scale function shown in FIG. - Only intervals up to are used, and the remaining intervals are zero.

Therefore, in the case of scale down, the scale function is applied to the filter coefficients read out from the coefficient memory. - Multiply the window function to limit the interval to. By the way, the window function - When expressed as a pulse present in the interval up to, since aliasing occurs in the scaled down image, as shown in FIG. 5 (b), a window function of decreasing size toward a peripheral part is used. An example of such a window function is a Gaussian function.

The window function is sampled the same as the scale function. That is, they have the same sampling period and number. The address for accessing the window memory is also determined by equation (2) at scale down.

5 is a diagram showing the effect of the window function in the case of scale down, FIG. 5 (a) shows the scale function used for scaling, FIG. 5 (b) shows the window function, and 5 (c) shows the scale function for scale down resulting from the influence of the window function. By the window function as shown in Figure 5 it is possible to obtain an image with reduced aliasing.

Eight filter coefficients are read from the coefficient memory for all interpolation points and multiplied by a window function and stored in the filter coefficient memory in turn.

The contents stored in each address fc_address of the filter coefficient memory are as follows.

fc_address [0] = coeff_1 [0], coeff_1 [1], coeff_1 [2], coeff_1 [3], coeff_1 [4], coeff_1 [5], coeff_1 [6], coeff_1 [7]

fc_address [1] = coeff_2 [0], coeff_2 [1], coeff_2 [2], coeff_2 [3], coeff_2 [4], coeff_2 [5], coeff_2 [6], coeff_2 [7]

,

,

fc_address [m-1] = coeff_m [0], coeff_m [1], coeff_m [2], coeff_m [3], coeff_m [4], coeff_m [5], coeff_m [6], coeff_m [7]

Here, m is an integer of 1 to M, and coeff_m [0] means the first filter coefficient at the mth interpolation point.

Window functions can also be used to scale up. In the case of scale up, the window function as shown in FIG. - The filter coefficients present in the other sections become the mode "0", which serves to substantially reduce the number of filter coefficients.

3) Output the stored filter coefficients according to the interpolation points.

For each clock, the position of the interpolation point is obtained by referring to the scale factor M, and the filter coefficient for each interpolation point is output by using the index as the index. The output filter coefficients are multiplied by pixel values of pixels involved in interpolation to determine pixel values of interpolation points.

6 is a block diagram showing the configuration of an apparatus for performing a filter coefficient generation method according to the present invention. The apparatus shown in FIG. 6 includes a filter coefficient calculating unit 600 and a filter coefficient storing and outputting unit 700.

The filter coefficient calculator 600 includes a table address generator 602, a coefficient memory 604, and a window memory 606.

The filter coefficient calculating unit 600 receives respective scale factors H_factor and V_factor and their inverses H_INV_factor and V_IN_factor for the horizontal and vertical directions.

The table address generator 602 may provide an address signal and a window memory provided to the coefficient memory 602 according to the respective scale factors H_factor and V_factor in the horizontal and vertical directions and their inverses H_INV_factor and V_IN_factor. Generates an address signal provided to 604.

The coeff address is determined by Equation 2 above, and the window address is determined by Equation 3 above.

The coefficient memory 604 reads the filter coefficients stored therein according to an address provided by the table address generator 602 and multiplies the window coefficients provided by the window function memory 606 to output them.

The window memory 606 reads and outputs window coefficients stored therein according to the address provided by the table address generator 602 (window address 0).

The filter coefficients output from the coefficient memory 604 are stored in the horizontal / vertical filter coefficient memories 704 and 708.

A signal RAM_sel for selecting the horizontal / vertical filter coefficient storage units 704 and 708 and a table address of the horizontal / vertical filter coefficient storage units 704 and 708 are generated.

The filter coefficient storage and output unit 700 includes a horizontal address generator 702, a horizontal filter coefficient storage unit 704, a vertical address generator 706, and a vertical filter coefficient storage unit 708. The filter coefficient storage and output unit 700 outputs the filter coefficients H_coeff and H_uv_coeff stored in the horizontal filter coefficient storage unit 704 only in the case of the horizontal synchronization signal H_SYNC and the horizontal active area H_START during horizontal scaling. In the vertical scaling, the filter coefficient V_coeff stored in the vertical filter coefficient storage unit 708 is output only when the vertical synchronization signal V_SYNC and the vertical active region V_START are enabled.

Here, H_uv_coeff are coefficients for the UV signal in the horizontal direction in which a U, V signal having a 4: 2: 2 format in the YUV signal system is scaled at a magnification of H_FACTOR * 2. When the coefficient table of H_FACTOR * 2 magnification is determined, it becomes the even-numbered address of the coefficient table of H_FACTOR * 2 magnification, and the coefficient table is used for both RGB and YUV signals.

The address provided to the horizontal filter coefficient formal unit 704 and the vertical filter coefficient storage unit 708 as a result of adding the input 1 / scale factor to each clock clk as a result of the horizontal address generator 702 and the vertical address generator 702. Generate a signal.

The operation of the apparatus shown in FIG. 6 will be described in detail.

1) Scale up process

(1-1) Scale up preparation process

Filter coefficients in the horizontal / vertical direction for the respective interpolation points are obtained from the coefficient memory 604 and stored in the horizontal / vertical filter coefficient storages 704 and 708.

(1-2) Scale up process

The filter coefficients stored in the horizontal / vertical filter coefficient storage units 704 and 708 are read out to perform scaling.

2) scale down process

(2-1) Scale Down Preparation

Obtain the filter coefficients in the horizontal / vertical direction for each interpolation points read out from the coefficient memory 604, and multiply them by the window coefficients read out from the window memory 606 to obtain the actual filter coefficients in the horizontal / vertical direction. This is stored in the horizontal / vertical filter coefficient storage units 704 and 708.

(2-2) Scale down process

The filter coefficients stored in the horizontal / vertical filter coefficient storage units 704 and 708 are read out to perform scaling.

FIG. 7 is a flowchart illustrating a scale up / down preparation process in the apparatus illustrated in FIG. 6, in which eight filter coefficients are used. 7 shows an offset determination process 702, a scale up / down determination process 704, a first coefficient memory address generation process 706, a coefficient memory access process 708, and a second coefficient memory address generation. A process 710, a window memory address generation process 712, a window memory access process 714, a coefficient generation process 716, and a coefficient storage process 718.

The scale up preparation process includes an offset determination process 702, a scale up / down decision process 704, a first coefficient memory address generation process 706, a coefficient memory access process 708, a coefficient generation process 716, and This is performed through the coefficient storage process 718.

Meanwhile, the scale down preparation process includes an offset determination process 702, a scale up / down determination process 704, a second coefficient memory address generation process 710, a window memory address generation process 712, a coefficient memory access process ( 708, the window memory access process 714, the coefficient generation process 716, and the coefficient storage process 718.

In FIG. 7, offset [N] represents an offset address at which the nth interpolation point is located in the coefficient memory 602, Factor represents the scale factor M, and INVFactor represents the inverse of the scale factor M. In FIG.

FIG. 8 is a flowchart illustrating a filter coefficient output process in the apparatus of FIG. 6. 8 includes an offset initialization process 802, a filter coefficient read process 804, a filter coefficient storage process 806, and a filter coefficient output process 808.

In FIG. 8, offset [N] represents an offset address at which the nth interpolation point is located in the coefficient memory 602, Factor represents a scale factor M, and INVFactor represents an inverse of the scale factor M. In FIG.

The filter coefficient generating method according to the present invention stores the filter coefficients for each interpolation point in the filter coefficient memory before scaling, and accesses the filter coefficient memory to output the filter coefficients at the time of scaling, so that the filter coefficients can be quickly generated. There is an effect that can be improved.

In addition, since the filter coefficients are generated by the memory access without generating the filter coefficients by the operation, the burden on the hardware is reduced.

Claims (3)

A filter coefficient generation method for image scaling for converting and outputting a resolution of an input image according to a predetermined scale factor M, Sampling a scale function for scaling and storing the scale function in a coefficient memory; Sampling a window function for limiting the scale function and storing the sampled window function in a window memory; Reading filter coefficients for each interpolation point from the coefficient memory with reference to a scale factor M; Reading window coefficients from a window memory with reference to the scale factor M; Storing filter coefficients obtained by multiplying the filter coefficients by a window coefficient in a filter coefficient memory; And And reading out and outputting filter coefficients stored in the filter coefficient memory for each interpolation point. The method of claim 1, wherein the storing of the scale function comprises storing only one scale function about the origin when the scale function is symmetric about the origin. The method of claim 1, wherein a sampling period and a sampling frequency for sampling the scale function and the window function in the scale function storing step and the window function storing step are the same.