JP2862644B2 - Image processing device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image processing for performing edge enhancement on a frequency image generated by orthogonal transformation.

[Conventional technology]

After performing a predetermined operation in the frequency domain, the frequency image compressed by the orthogonal transformation operation is expanded by the orthogonal inverse transformation operation and restored to the original image before the orthogonal transformation. There is a problem that high-frequency components of the original image are lost during the compression and decompression processes. Compensate for the lost high frequency components,
Enhance image contours to improve apparent resolution. This is called outline emphasis.

Conventional edge enhancement uses an enhancement filter that operates in real time using an analog filter using an analog delay line, and a digital filter that improves delay time fluctuation, S / N, and linearity of the delay line by digitizing. An addition circuit configuration and a method using two-dimensional convolution are known.

[Problems to be solved by the invention]

However, these methods are disadvantageous in terms of hardware technology because the circuit configuration for contour enhancement is complicated and large.

SUMMARY OF THE INVENTION The present invention has been made in view of the above, and provides an image processing apparatus that performs an outline enhancement of an image on a compressed frequency image using an orthogonal transformation operation with a circuit configuration advantageous in hardware technology. The purpose is to do.

[Means to solve the problem]

In order to achieve the above object, an image processing apparatus according to the present invention is an image processing apparatus for inversely orthogonally transforming a frequency image and converting it into a real image. Image data storage means, orthogonal transform component time series conversion means for converting a two-dimensional orthogonal transform function of frequency image data stored in the storage means into a time series of orthogonal transform components in a frequency domain, and the orthogonal transform component time series The orthogonal transformation component at a predetermined time point on the time series of the orthogonal transformation component converted into the time series by the conversion means is stored at a predetermined time on the time series other than the time point. A time series time conversion unit that outputs a conversion component, and an orthogonal transformation component whose time point on a time series is converted by the orthogonal transformation component time series conversion unit. An orthogonal transform component superimposing unit that superimposes the orthogonal transform component at the time of synchronization with the time on the time series of the orthogonal transform component converted by the time series time transform unit; and an orthogonal transform component superimposed by the orthogonal transform component superimposing unit. And real screen data generating means for generating real image data by subjecting the frequency image data to inverse orthogonal transform.

Further, a predetermined constant corresponding to each orthogonal transform component at the predetermined time is stored, a constant storage output means for outputting the constant in synchronization with the other time, and a constant output by the constant storage output means. And a multiplier for generating a product of the orthogonal transform component output at the other point in time.

Further, the predetermined constant is a non-negative number less than 1 which is set in advance.

A memory storing the predetermined non-negative number less than 1;
A memory for storing the predetermined constant that can be converted by the operator in a non-negative range less than 1.

In addition, the apparatus further includes a constant conversion unit that allows the operator to convert the predetermined number to a non-negative number less than 1.

Further, the time series transform by the orthogonal transform component time series transforming means is performed by zigzag scanning the AC orthogonal transform coefficients of the inversely quantized orthogonal transform coefficients block by block.

Further, the orthogonal transform is a discrete cosine transform (DCT).

〔Example〕

FIG. 1 is a functional block diagram of an embodiment of the present invention.
Reference numeral 1 denotes a rewritable memory card in which, for example, an SRAM semiconductor memory is in the form of an integrated circuit (IC) card, which stores frequency image data compressed by performing an adaptive discrete cosine transform operation on spatial image data. 2 is a memory card 1
Is a Huffman decoder prepared for the output 10 of FIG. The output 11 of the Huffman decoder 2 is connected to the inverse quantizer 3.
Numeral 4 is connected to the inverse quantizer 3 and a random access memory (RAM) for storing predetermined data 12 out of the frequency image data, and numeral 5 is connected to an output 13 of the random access memory (RAM) 4 and its output. This is a data converter that combines the data 13 with the output data 12 of the inverse quantizer 3. Reference numeral 6 denotes an inverse discrete cosine converter for performing an inverse discrete cosine transform operation on the converted data 16 converted by the data converter 5, and reference numeral 7 denotes a digital / analog converter connected to an output 17 of the inverse discrete cosine converter 6. Reference numeral 8 denotes an output terminal of the digital / analog converter 7. 9 is random access memory (RAM)
4 is connected to the data converter 5 and controls the address of the random access memory (RAM) 4, and controls the conversion of orthogonal transform (discrete cosine transform) coefficients to the data converter 5. is there.

The operation of the present invention in such a configuration will be described.
The compressed image data stored in the memory card 1 is read out for each small block of the divided image, and the output is output.
10 is decoded by the Huffman decoder 2 for each small block, the decoded output 11 is inversely quantized by the inverse quantizer 3 and the code is restored, and the output 12 is obtained by the discrete cosine transform (Discrete Cosine Transform; DCT)
Is generated as frequency image data. The DCT is said to have a higher compression efficiency when compressing image data represented as a spatial light distribution in one form of orthogonal transform, as compared with transforms belonging to other forms of orthogonal transform.

An image composed of N × N pixels is converted into a two-dimensional discrete signal using two transforms m and n, and f (m, n), m, n
Are represented by 0, 1,..., N−1, and the transform coefficient of the two-dimensional DCT of f (m, n) is given by equation (1).

In the above formula v is a spatial frequency in the horizontal direction and the vertical direction of the target spatial image, respectively.

FIG. 2 shows the frequency image data whose code has been restored by inverse quantization by the inverse quantizer 3 as a two-dimensional 8 × 8 pixel DCT function. The data is frequency image data generated by performing DCT on one block of data obtained by dividing a spatial image for one screen into small blocks. The size of a block, which is a unit of conversion, is divided into small image blocks (N × N) consisting of predetermined pixels (N pixels) set in the horizontal and vertical directions based on the orthogonal conversion efficiency. Is performed, and a DCT coefficient is obtained.
At this time, the size of N is set to 8 to 16 based on the DCT conversion efficiency. As N increases, the conversion efficiency improves, but the tendency of the improvement shows a saturation tendency and the scale of the device increases as the scale of operation increases.
In this embodiment, N is set to 8. 2D 8x
A two-dimensional DCT is performed for each small block composed of eight pixels, and a DCT coefficient is obtained for a two-dimensional frequency, which is expressed as a two-dimensional 8 × 8 pixel DCT function. The spectral information of the spatial image is extracted from the data expressed as a result of executing the two-dimensional DCT. The spectrum information is so-called spatial frequency spectrum information, which expresses a feature of the image, that is, a certain periodic component or a synchronizing structure in the image.

If N = 8 in the DCT equation (1), the equation (1) becomes the equation (2).

F (u, v) in the expression (2) can be obtained by performing DCT in the m direction using f (m, n) and n as constants, and then performing the same transformation on n. This conversion focuses on the periodicity of the two-dimensional spatial image, converts digital data corresponding to the discrete angular velocity of the Fourier transform converted to the frequency axis, and converts only digital data using cosine coefficients.
If N = 8, a two-dimensional discrete cosine function of 8 × 8 pixels (one block) is shown as in FIG. In the figure, u is a spatial frequency in the horizontal direction of the block when the area of the spatial image is divided into blocks of 8 × 8 pixels and two-dimensional DCT is performed for each block. u is the spatial frequency in the vertical direction of the block when the DCT is executed. The position on the frequency plane corresponding to each of the spatial frequencies u and v is represented by a number from 1 to 64. Each of these numbers has an address function on a two-dimensional frequency plane with respect to the frequency component distribution at the position on the plane, that is, the DCT coefficient distribution. In the figure, the DCT coefficient at the position specified by the address 5 is the DCT coefficient at the horizontal spatial frequency u2 specified by the address 2 and the DCT coefficient at the vertical spatial frequency v3 specified by the address 3
The synthesized spectrum DCT coefficient F (u2, v
3) is expressed. The composite spectrum DCT coefficient F
By transforming (u, v), it is possible to emphasize image quality features in the spectral domain. This is a feature of the orthogonal transform method. The data converter 5, the random access memory (RAM) 4, and the conversion controller 9 perform the conversion for F (u, v) on the u, v plane (spectral domain) shown in FIG.

The conversion will be described with reference to FIGS. 1 to 3 and FIG. FIG. 3 shows the data converter 5 shown in FIG.
Random access memory (RAM) 4 and conversion control unit 9
Is a block circuit obtained by further functionalizing the functional block composed of. In FIG. 3, the conversion control unit 9 mainly has a counter. Reference numeral 30 denotes a constant generator which receives a read pulse generated by the conversion control unit 9 having a counter and generates a constant to be multiplied by the output 13 of the random access memory (RAM) 4 and is constituted by a read only memory (ROM). The constant is a non-negative number less than 1 which is predetermined in consideration of the degree of contour enhancement. 31 is a multiplier for performing multiplication of the output 13 of the random access memory (RAM) 4 and the output (constant) of the constant generator 30, and 33 is a multiplier 33
This is an adder for adding the output (product) 34 of the data and the data 12.

The conversion operation will be described under such a configuration. The frequency image data 12 that has been subjected to DCT is a scan output obtained by zigzag scanning for each block. FIG. 6 shows the structure shown in FIG.
FIG. 9 illustrates a state in which zigzag scanning is performed on AC orthogonal transform coefficients of orthogonal transform coefficient data for blocks. The area of the hatched portion 60 at the upper left corner corresponds to the output of the DC orthogonal transform coefficient, and the other area corresponds to the output of the AC orthogonal transform coefficient. In the zigzag scan, the coefficients dequantized along the scan 61 are executed from a low-frequency component to a high-frequency component. The scan output (frequency image data) 12 is a random access memory (RA
M) 4 and one of the addition inputs of the adder 33.
The counter constituting the conversion control unit 9 is supplied with a clock pulse CP synchronized with the sequence of DCT coefficient generation in zigzag scanning and a block start pulse BP for instructing the start of each block. Random access memory (RAM)
A predetermined address on a zigzag scan corresponding to a DCT coefficient to be transformed to contribute to contour emphasis among 64 DCT coefficients in a sequence from 1 to 64 for each block supplied to 4 is preset. Keep it. In the present embodiment, the predetermined address is 7, 10, 1
Set to 6, 21, 25, 52. When converting the DCT coefficient corresponding to the address 7, the DCT coefficient corresponding to the address 2
The product of the coefficient multiplied by a predetermined constant is the DC corresponding to address 7.
The converted DCT coefficient at the address 7 is obtained by adding to the T coefficient. In general terms, when a DCT coefficient corresponding to an address Q (hereinafter, referred to as DCTCQ) is converted, a DCT coefficient (hereinafter, referred to as DCTCP) corresponding to an address P is converted into a predetermined constant. A product (K.DCTCP) multiplied by K (K will be further described later) is added to DCTCQ to obtain a converted DCT coefficient at the address Q (hereinafter, this is represented by DCTC (Q ← P)). This is represented by an equation: DCTC (Q ← P) = DCTCQ + K.DCTCP (3) A set of addresses of Q ← P is set in advance. In this embodiment, a set of Q ← P addresses is set as follows. 7 ← 2, 10 ← 3, 21 ← 4, 25 ← 5, 16 ← 6, 52 ← 1
3. In such a setting mode, the counter belonging to the conversion control unit 9 is reset by the arrival of the block start pulse BP, and the counter comes from 1 to 64 in the zigzag scan by the next clock pulse CP.
It is counted in synchronization with the sequence of the T coefficient. In this process, the conversion control unit 9 counts 2,
Random access memory (RA
M) Output the write command 15 to 4. The random access memory (RAM) 4 which has received the write command 15 writes the DCT coefficient DCTCP corresponding to the address (2, 3, 4, 5, 6) on the zigzag scanning line corresponding to each count number into the zigzag scanning input 12. Select and write from the sequence. In addition, the conversion control unit 9 counts 7, 10, 16, 21, 2
In steps 5 and 52, a read command 14 is output to the random access memory (RAM) 4. The random access memory (RAM) 4 which has received the read instruction 14 stores the previously written DCT.
The coefficient DCTCP (P = 2, 3, 4, 5, 6) is converted into the address Q (Q = 7, 10, 16, 21, 2, 2) in the address set Q ← P.
5, 52), that is, at the count Q by the clock pulse CP. The read output 13 is supplied to the multiplier 31 in synchronization with the clock pulse CP. On the other hand, the read command 14 for the random access memory (RAM) 4 is supplied to the constant generator 30 in synchronization with the clock pulse CP and functions as the constant read command 14. The constant generator 30 is constituted by a read-only memory, and a constant K constituted by a non-negative number less than 1 is previously set and stored in the memory. The value of the constant K is associated with DCTCP in the equation (3), and corresponds to each P (P = 2, 3, 4, 5, 6, 13), for example, K = 1/2, 1/3, 1/4, 1/5, 1 /
It is preset as 6, 1/13 or 1/2, 1/9, 1/16, 0, 1/36, 1/124, etc. The read command 14 is P =
A constant K multiplied by DCTCP corresponding to 2, 3, 4, 5, 6, 13 is read from the constant generator 30 and a read output (constant K) 32 is transferred to the multiplier 31. The multiplier 31 transfers the K DCTCP in the equation (3) as a sequence output 34 for each K and P to the other addition input of the adder 33 in synchronization with the clock pulse CP. The adder 33 outputs the sequence output 34
And the zigzag scanning output 12 supplied to one of the addition inputs, and the addition output 16 is used as the output of the data
Transfer to T unit 6. The addition output 16 is obtained by performing the operation on the right-hand side of the equation (3) in the adder 33 so that the DCTC (Q ←
P) is configured as one block of sequence data that includes P) in the sequence. The data is data on which outline enhancement has been performed on one block. The counter of the conversion control unit 9 is reset by the block start pulse BP of the next block, and the added output 16 is generated by the subsequent clock pulse in the same manner as described above. By similarly generating the addition output 16 for all the blocks for one screen, the data on which the outline enhancement has been performed for one screen is transferred to the inverse DCT unit 6. The inverse DCT unit performs inverse DCT on the DCT data on which contour enhancement has been performed for each block to image data for each block. When the inverse DCT of all blocks is completed in this way, all blocks are integrated and an image is restored. The digital image 17 is converted into an analog signal by the digital / analog converter 7 and output from the output terminal 9 as a video signal.

FIG. 4 is a diagram showing another embodiment in which the functional blocks including the data converter 5, the random access memory 4, and the conversion control unit 9 shown in FIG. 1 are further functionalized. In this embodiment, the conversion control unit 9 in the embodiment shown in FIG.
And a constant conversion control unit 42. The constant conversion data input means 41 is, for example, a keyboard. keyboard
41 is connected to a constant conversion control unit 42, and the output of the control unit 42
44 is supplied to the constant generator 30. The storage unit of the constant generator 30 is configured as a rewritable memory, for example, a random access memory (RAM). The constant K of the constant generator 30 preset in correspondence with each P (P = 2, 3, 4, 5, 6, 13) in relation to DCTCP in the equation (3) is inputted by the operator from the keyboard 41. The constant data and the control data are converted under the conversion control of the constant conversion control unit 42, and the operator can change the form of the outline emphasis. Other operations for generating the contour-emphasized addition output 16 are the same as those in the embodiment shown in FIG.

FIG. 5 is a diagram showing another embodiment in which the functional blocks including the data converter 5, the random access memory 4, and the conversion control unit 9 shown in FIG. 1 are further functionalized. In this embodiment, the conversion control unit 9 in the embodiment shown in FIG. 4 is constituted by a constant generator 50 in addition to the counter 40, the constant conversion data input means 41, and the constant conversion control unit 42. The storage unit of the constant generator 30 is configured by a read-only memory (ROM), and the storage unit of the constant generator 50 is a rewritable memory, for example, a random access memory (RA).
M). The constant generator 50 has an input side connected to the constant conversion control unit 42, an output side connected to the adder 31, and receives control of the read pulse 4 output from the counter 40. The constant K of the constant generator 50 is converted by the constant data and control data input by the operator from the keyboard 41 under the conversion control of the constant conversion control unit 42, and the operator can change the form of the outline emphasis. it can. Constant K of constant generator 30
Is associated with DCTCP in equation (3), and each P (P = 2,
3, 4, 5, 6, 13) are set in advance. The read pulse 14 output from the counter 40 is output to the constant generators 30 and 50 by a constant K set for each generator.
Functions as a read instruction 14 for outputting the same to the multiplier 31. The K constant outputs 32 and 51 of the generators 30 and 50 synchronously read by the read instruction 14 are multiplied and superimposed in the multiplier 31 to contribute to the contour emphasis mode. Expands. Other operations for generating the contour-emphasized addition output 16 are the same as those in the embodiment shown in FIG.

〔The invention's effect〕

The present invention relates to an adder circuit using a digital filter for emphasizing a feature (contour) of image quality by changing a specific transform output on a frequency spectrum using a feature of an orthogonal transform coding method as described above. Alternatively, since the configuration can be realized without a configuration such as a circuit using two-dimensional convolution, an excellent effect is obtained in the circuit configuration.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a functional block diagram of an embodiment of the present invention, and FIG. 2 is a diagram expressing a discrete cosine function of an image composed of two-dimensional 8 × 8 pixels. 3 to 5 are functional block diagrams showing a configuration in which the conversion function centering on the data converter of FIG. 1 is further functionalized, and FIG.
FIG. 3 is a zigzag scanning state diagram of AC orthogonal transform coefficients showing a state in which the discrete cosine function expressed in FIG. 2 is zigzag scanned. 1 ... Memory card, 2 ... Huffman decoder, 3 ... Dequantizer, 4 ... Random access memory (RAM), 5
…… Data converter, 6 …… Inverse discrete cosine converter (Inverse DC
T unit), 7 ... Digital / analog converter, 9 ... Conversion control unit, 30, 50 ... Constant generator, 31 ... Multiplier, 33 ...
Adder, 40 counter, 41 constant data input means (keyboard), 42 constant conversion control unit, 61 zigzag scanning line.

Claims (7)

(57) [Claims] 1. An image processing apparatus for inversely orthogonally transforming a frequency image into a real image, comprising: frequency image data storage means for storing frequency image data that has been orthogonally transformed and compressed; and frequency image data stored in the storage means. Orthogonal transform component time series conversion means for converting a two-dimensional orthogonal transform function of the frequency image data into a time series of orthogonal transform components in the frequency domain; and an orthogonal transform component converted into a time series by the orthogonal transform component time series transform means. A time-series time conversion unit that stores an orthogonal transformation component at a predetermined time point set in advance on the time series, and outputs the conversion component at another time point set in advance on the time series excluding the time point; The orthogonal transformation component whose time point on the time series is transformed by the orthogonal transformation component time series transformation means is converted to the direct transform time converted by the time series time transformation means. Orthogonal transform component superimposing means for superimposing on the orthogonal transform component at the time of synchronization with the time on the time series of the intersecting transform component, and inverse orthogonal transform of the frequency image data generated by being superimposed by the orthogonal transform component superimposing means. An image processing apparatus, comprising: real screen data generating means for generating real image data. 2. A constant storage output means for storing a predetermined constant corresponding to each orthogonal transform component at the predetermined time, outputting the constant in synchronization with the other time, and outputting by the constant storage output means. The image processing apparatus according to claim 1, further comprising: a multiplier configured to generate a product of the set constant and the orthogonal transform component output at the other time. 3. The method according to claim 1, wherein the predetermined constant is a predetermined one.
The image processing apparatus according to claim 2, wherein the non-negative number is less than. 4. The constant storage output means stores a predetermined non-negative number less than 1 and a predetermined constant which can be converted by an operator in a non-negative range less than 1. The image processing apparatus according to claim 2, further comprising: 5. The image processing apparatus according to claim 2, further comprising a constant conversion unit that allows an operator to convert the predetermined number to a non-negative number less than 1. 6. The time series transform by said orthogonal transform component time series transforming means is performed by zigzag scanning an AC orthogonal transform coefficient of an inversely quantized orthogonal transform coefficient for each block. The image processing apparatus according to claim 3. 7. The orthogonal transform is a discrete cosine transform (DCT).
The image processing apparatus according to claim 5, wherein