JP4414328B2 - Image processing apparatus and image processing apparatus - Google Patents

Description

  The present invention relates to an image processing apparatus capable of embedding a large amount of additional information in quantized frequency image data and a related technique.

  In recent years, digital contents that are digitized audio and video data are increasing. Since digital content can be easily copied exactly the same as the original, protecting the copyright of the content is an important issue. Since illegally copied and distributed content is indistinguishable from the original, it is difficult to provide evidence that claims the copyright of the content, and copyright protection methods are being studied.

  “Digital watermark” is used as a method for protecting the copyright of digital contents. Digital watermarking is a technology that embeds data in audio or video data so that it is difficult for the viewer to perceive, and extracts the embedded data from the embedded data. In order to protect the copyright, copyright information such as a copyright holder name and date is embedded as a digital watermark in audio or video data. The copyright information embedded from the illegally copied content is detected, the copyright owner of the content is identified, and illegal copying is prevented. If information is embedded in the entire content, it is possible to specify not only the presence / absence of alteration but also the specific alteration location.

  Several methods have been proposed in which additional information is embedded in encoded image data compressed in accordance with an international standard such as JPEG / MPEG and the additional information is extracted during decoding.

  For example, Patent Document 1 (Japanese Patent Laid-Open No. 2002-330279) discloses a method for embedding data in an image and a method for extracting the data.

  At the time of embedding, the original image is divided into 8 × 8 pixel block units. All block units constituting the entire screen are subjected to DCT, and frequency image data (here, DCT coefficients) is generated. Each block is quantized using a quantization table, and one coefficient belonging to the highest frequency in the quantized frequency image data is replaced with 1-bit data. By this replacement, the data to be embedded is embedded in the quantized frequency image data. Furthermore, run-length encoding and Huffman encoding are performed in sequence, and a value corresponding to one coefficient belonging to the highest range among the values in the quantization table is replaced with “1”.

  A device such as a digital camera or a color facsimile using JPEG encoding compresses and encodes an input image by hardware. Therefore, adding a process of embedding a digital watermark before compression encoding increases the cost significantly. To avoid this, the encoded data or quantized frequency image data is input, and the digital watermark information is input to the input data based on the table used for encoding the input data or the table used for quantization. Just embed.

  However, when embedding a large amount of additional information of several hundred bytes or more in image-compressed image-compressed data, according to Patent Document 1, a desired number of frequency image data to be embedded (generally, generally, It is necessary to replace the data with 1 bit only (not “1”).

  By the way, the DCT coefficient of the high-frequency component often becomes “0” by quantization, and a “0” lump that often reaches the end of the high-frequency component appears. Using this property, in the encoding / decoding process, on the zigzag scan path starting from the direct current component to the end of the high frequency component, a block of “0” reaching the end of the high frequency component is converted to EOB (End of Block) is replaced with a unique code. By doing so, several to several tens of “0” chunks can be replaced with only one code, and the amount of codes can be significantly reduced.

  However, in the conventional technique, when one coefficient belonging to the highest range is replaced with “1”, the end of the high frequency component becomes “1”, and therefore there is a “0” block leading to the end of the high frequency component. Will not. Therefore, replacement by EOB cannot be performed, and as a result, the amount of codes is greatly increased.

  In addition, it is not impossible to replace the DCT coefficients of the middle and low frequency components. However, the reason why one coefficient belonging to the highest range is replaced in the above-described technique is that image quality degradation is considered to be small even if this coefficient is replaced. If the DCT coefficients of the middle and low frequency components are replaced, these coefficients are deeply related to the image quality and have a relatively large value, so that the image quality deterioration cannot be avoided.

  Japanese Patent Laid-Open No. 2000-151973 discloses a technique for embedding a large amount of additional information in image data. Now, it is desirable that the sum of the absolute values of the differences between the frequency image data is as small as possible before and after the information is embedded. This is because the fact that the total sum is large means that the image has changed greatly, in other words, the image quality of the image in which the information is embedded is significantly deteriorated as compared with the image before embedding.

The inverse quantization process is basically a process of multiplying the corresponding value of the quantization table by the corresponding value of the quantized frequency image data. In general, when the compression rate is relatively high, the value of the quantization table is very large. Therefore, according to Patent Document 2, even if the value of the quantized frequency image data is slightly manipulated to embed information (for example, “+1”, etc.), if the compression ratio is relatively high, Since the value is very large (for example, “100”), only a few operations are amplified several times to hundred times or more (difference: + 1 * 100 = + 100), and the image quality of the decoded image is obtained by embedding information. There is a problem that the quality of the material deteriorates significantly.
JP 2002-330279 A JP 2000-151973 A

  Therefore, an object of the present invention is to provide an image processing apparatus that can embed a large amount of additional information while suppressing deterioration in image quality even for input data encoded at a high compression rate.

  An image processing apparatus according to a first invention uses a first quantization table, a second quantization table different from the first quantization table, and a first quantization using the first quantization table. An inverse quantization unit that inversely quantizes the generated data to generate frequency image data, a quantization unit that quantizes the frequency image data using a second quantization table, and generates second quantized data; An information embedding unit that embeds additional information in the second quantized data.

  In the image processing apparatus according to the second invention, in the second quantization table element, at least the value of the element belonging to the part where the additional information is embedded is the part where the additional information is embedded in the element of the first quantization table. Is smaller than the value of the element belonging to.

  With the above configuration, the operation for embedding additional information can be prevented from being amplified by a large value in the quantization table. Therefore, a large amount of additional information can be embedded in input data encoded at various compression rates while suppressing deterioration in image quality.

  According to the present invention, a large amount of additional information can be embedded while maintaining the image quality of compression-encoded input data.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the present invention. As shown in FIG. 1, the image processing apparatus according to the present embodiment includes a decoding unit 101, an inverse quantization unit 102, a quantization unit 103, an information embedding unit 104, and an encoding unit 105. The image processing apparatus further includes a first quantization table 106, a second quantization table 107, a table control unit 108, and an embedded information storage unit 109 that supplies embedded information to the information embedding unit 104.

  The decoding unit 101 receives a bit stream, decodes the bit stream, and generates quantized frequency image data (DCT coefficients). The inverse quantization unit 102 inversely quantizes the DCT coefficient quantized using the first quantization table 106 illustrated in FIG. 3 to generate a DCT coefficient. For example, when the quantized DCT coefficients are as shown in FIG. 7A and the contents of the first quantization table 106 are as shown in FIG. Such a DCT coefficient is output.

  The quantization unit 103 quantizes the DCT coefficient using the second quantization table 107 as illustrated in FIG. 4 or FIG. 5 to generate a quantized DCT coefficient. For example, when the DCT coefficient is as shown in FIG. 8A and the content of the second quantization table 107 is as shown in FIG. 8B, the quantization as shown in FIG. The obtained DCT coefficient is output.

  As will be described in detail later, in this embodiment, the relationship between the first quantization table 106 and the second quantization table 107 is controlled by the table control unit 108. However, when both the first quantization table 106 and the second quantization table 107 can be fixedly determined in advance, the table control unit 108 may be omitted.

  The information embedding unit 104 inputs the embedding information from the embedding information storage unit 109 and embeds the information in the quantized DCT coefficient output from the quantization unit 103 according to a rule described later.

  The encoding unit 105 encodes the quantized DCT coefficient in which information is embedded by the information embedding unit 104, and generates a bit stream.

  Next, as shown in FIG. 10A, the case where the embedded information (13, 46) is embedded will be described in detail. Here, it is assumed that the first quantization table 106 has the contents shown in FIG.

  As shown in FIG. 6, the table control unit 108 acquires the embedding information (13, 46) from the embedding information storage unit 109 in step 300 and binarizes it. As a result, a bit string having a 16-bit length of (0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 0, 0, 1, 1, 0) is obtained. In step 301, the table control unit 108 determines the position of the target coefficient involved in information embedding by this binarization. As indicated by an arrow N in FIG. 10A, in this example, 16 target coefficients are determined along the zigzag scan path starting from the DC component.

  In step 302, the table control unit 108 obtains the target coefficients (16, 11, 12, 14,..., 24, 40) of the first quantization table 106 shown in FIG. Then, the square sum S is obtained. In step 303, the table control unit 108 copies the contents of the first quantization table 106 as they are to the second quantization table 107.

  In step 304, the table control unit 108 compares the square sum S with a preset threshold value TH. If S> TH, it can be said that the value of the second quantization table 107 is relatively large and the compression rate is high. At this time, in step 305, the table control unit 108 obtains a multiplication value K (0 <K <1). The multiplication value K may be a predetermined constant value (for example, K = 0.5) or may be determined based on the square sum S. In step 306, the table control unit 108 multiplies each value in the second quantization table 107 by the multiplication value K. Since 0 <k <1, the absolute value of the multiplication result is naturally smaller than that before the multiplication.

  When all the values in the first quantization table 106 shown in FIG. 3 are multiplied by the multiplication value K = 0.5, the second quantization table 107 becomes as shown in FIG. However, since only the portion corresponding to the target coefficient is involved in the information embedding, as shown in FIG. 5, only a part of the values in the first quantization table 106 is multiplied by the multiplication value K. Also good.

  In step 304, when the sum of squares S is less than or equal to the threshold value TH, it can be said that the value of the second quantization table 107 is relatively small and the compression rate is low. At this time, in this embodiment, the table control unit 108 uses the second quantization table 107 as it is without changing the value. As a result, the same quantization table as the first quantization table 106 is used. However, even at this time, a larger multiplication value K (for example, K = 0.9) can be multiplied.

  Next, the processing of the information embedding unit 104 will be described with reference to FIG. When the information embedding unit 104 receives the quantized DCT coefficient from the quantizing unit 103, in step 400, the information embedding unit 104 acquires the embedding information from the embedding information storage unit 109 and binarizes it. In step 401, the information embedding unit 104 obtains the bit length L of the binarized embedded information, initializes the counter i to “1”, and the information embedding unit 104 performs the bit length L in step 402 and subsequent steps. As a result, the embedded information is embedded in the quantized DCT coefficient.

  That is, in step 403, the information embedding unit 104 sets the i-th bit value of the embedded information to the variable Val, and performs a zigzag scan on the quantized DCT coefficient (the head is a DC component; FIG. 10A arrow N The i-th coefficient of (see) is set in the variable C.

  In step 404, the process branches depending on the variable Val. When the variable Val is “0”, the process proceeds to step 405, and when the variable Val is “1”, the process proceeds to step 409.

  In step 405, the information embedding unit 104 checks whether the variable C is an odd number. If the variable C is an even number, the information embedding unit 104 shifts the processing to step 413. If it is an odd number, it is checked in step 406 whether the variable C is positive. If it is positive, “1” is subtracted from the variable C in step 408, and if negative, “1” is added to the variable C in step 407. As a result, the variable C approaches “0”.

  In step 409, the information embedding unit 104 checks whether or not the variable C is an even number. If the variable C is an odd number, the process proceeds to step 413. If it is an even number, it is checked in step 410 whether the variable C is positive. If it is positive, “1” is subtracted from the variable C in step 411, and if negative, “1” is added to the variable C in step 412. As a result, the variable C approaches “0”.

  In step 413, the information embedding unit 104 adds “1” to the counter i, and repeats the processing from step 402.

  As described above, as shown in FIG. 10A, for example, when the embedded information is (13, 46), as a result of binarization, (0, 0, 0, 0, 1, 1, A bit string having a 16-bit length of 0, 1, 0, 0, 1, 0, 0, 1, 1, 0) is obtained. In this case, since the first 3 bits of the bit string are all “0”, in the DCT coefficients (32, 9, -14) corresponding to these bits, the DCT coefficient (32) is an even number and is not changed. Since the next DCT coefficient (9) is an odd number, it is changed to an even number “8” close to “0”. Further, since the next DCT coefficient (−14) is an even number, it is not changed. Similarly, processing by the information embedding unit 104 is performed, and the result is as shown in FIG. Note that the hatched rectangle in FIG. 10B is a target coefficient.

  Next, the overall processing by the image processing apparatus of this embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing processing performed by the image processing apparatus of FIG.

  First, the decoding unit 101 performs entropy decoding such as Huffman decoding on the input compressed encoded data to generate first quantized data (step S201). In general, in the case of a still image, the input compressed encoded data is JPEG data.

  Next, the inverse quantization unit 102 inversely quantizes the first quantized data output from the decoding unit 101 based on the first quantization table 106 used at the time of quantization to generate frequency image data ( Step S202). Here, since the frequency image data is generated from JPEG data, it is a DCT coefficient.

  Next, the quantization unit 103 uses the second quantization table in FIG. 4 (different from the first quantization table 106), which is different from the first quantization table used in the inverse quantization unit 102. The frequency image data (DCT coefficient) is quantized to generate second quantized data (step S203). Note that each element of the second quantization table 107 has a smaller value than each element of the first quantization table 106.

  Next, the information embedding unit 104 embeds predetermined additional information by operating the second quantized data output from the quantizing unit 103 according to a predetermined rule (step S204). Note that the embedding method operated according to a predetermined rule may change the value of the quantized data to an even number or an odd number according to the bit value of the additional information as described above.

  Regarding the amount of embedded data, for example, when the input image size is 320 × 240 pixels and additional information of 16 bits (2 bytes) is embedded in a DCT block unit of 8 × 8 pixels, the input image composed of 320 × 240 pixels is 1200 DCT blocks can be used, and a total of 2400 bytes of information can be embedded.

  Next, the encoding unit 105 generates JPEG compressed data in which the additional information is embedded by entropy encoding the quantized data in which the additional information is embedded, such as a Huffman code (step S205).

  Next, referring to FIG. 11, the quantized DCT coefficient in which information is embedded and the first quantization table 106 having a relatively large value and a high compression rate by the image processing apparatus according to the present embodiment are stored. The case where the second quantization table 107 is used as it is will be specifically compared. First, it is assumed that the quantized DCT coefficient before embedding information has contents as shown in FIG.

  At this time, according to this embodiment, a result as shown in FIG. 11B is obtained. At this time, the sum of the absolute values of the DCT coefficient differences before and after embedding is “54”.

  On the other hand, in the same case, if the first quantization table 106 is directly processed as the second quantization table 107, a result as shown in FIG. 11C is obtained. At this time, the sum is “124”.

  That is, according to the present embodiment, the sum is reduced to 1/2 or less. In other words, it will be understood that, although the same information is embedded, according to the present embodiment, image quality deterioration can be significantly suppressed.

  As described above, according to the image processing apparatus of this embodiment, each element of the quantization table is converted into a small value for the data that has already been compressed and encoded with the quantization table having a large value. Since quantization is performed, even if the value of the quantized data is slightly changed to embed additional information, the value converted by inverse quantization does not greatly differ from the value before embedding. As a result, a large amount of additional information can be embedded while maintaining the image quality of the compression-encoded input data.

  In addition, since the high frequency component DCT coefficient of the data that has already been compressed with the large quantization table, which is the first quantization table, is almost zero, the second quantum table again. Even if it is quantized with the quantization table, it becomes zero as it is, and by embedding the additional information only in the quantized data corresponding to the middle and low frequency components, it is possible to reduce the increase in the code amount of the encoded data.

  Note that the input compressed encoded data may be compressed encoded data such as moving image encoded MPEG data or JPEG 2000 in addition to JPEG data. That is, the present invention is not limited to this as long as it is encoded data with quantization.

  Furthermore, as shown in FIG. 5, the second quantization table 107 does not have to be all elements smaller than the first quantization table 106, and the upper left corresponding to the quantized data in which the additional information is embedded It is sufficient that only the position value is smaller than the first quantization table 106.

  If the first quantization table 106 has a small value from the beginning, it is not necessary to replace the second quantization table 107, and the same effect can be obtained by using the first quantization table 106 as it is.

  Typically, each function constituting the image processing apparatus according to the present embodiment includes a storage device such as a ROM, a RAM, and a hard disk in which predetermined program data is stored, and a CPU (central processing) that executes the program data.・ Unit). In this case, each program data may be introduced via a storage medium such as a CD-ROM or a flexible disk.

  The image processing apparatus according to the present invention can be suitably used, for example, in the field of embedding information in data compliant with standards such as JPEG / MPEG.

1 is a block diagram of an image processing apparatus according to Embodiment 1 of the present invention. Flowchart of the image processing apparatus in Embodiment 1 of the present invention FIG. 3 is an exemplary diagram of a first quantization table in Embodiment 1 of the present invention. Exemplary diagram of second quantization table in Embodiment 1 of the present invention Exemplary diagram of second quantization table in Embodiment 1 of the present invention Flowchart of the table control unit in Embodiment 1 of the present invention (A) Exemplary diagram of quantized DCT coefficients in Embodiment 1 of the present invention (b) Exemplary diagram of first quantization table in Embodiment 1 of the present invention (c) Embodiment 1 of the present invention Of DCT coefficient (A) Illustration of DCT coefficient in Embodiment 1 of the present invention (b) Illustration of second quantization table in Embodiment 1 of the present invention (c) Quantization in Embodiment 1 of the present invention Of DCT coefficient Flowchart of information embedding unit in Embodiment 1 of the present invention (A) Explanatory drawing of the information embedding process in Embodiment 1 of this invention (b) The example figure of the information embedding result in Embodiment 1 of this invention (A) Exemplary diagram of DCT coefficient (before embedding) in the first embodiment of the present invention (b) Exemplary diagram of DCT coefficient (after embedding) in the first embodiment of the present invention (c) DCT coefficient (embedding) in the comparative example Illustration after)

Explanation of symbols

101 Decoding unit 102 Inverse quantization unit 103 Quantization unit 104 Information embedding unit 105 Encoding unit

Claims (7)

A first quantization table;
The first Unlike quantization table, and the value of elements belonging to at least portions of the additional information is embedded, the first value of belonging to subelements said additional information is embedded in the elements of the quantization table A second quantization table smaller than ,
And inverse quantization unit for generating frequency image data and first inverse quantization quantum cade over data that has been quantized using said first quantization table,
A quantization unit that quantizes the frequency image data using the second quantization table and generates second quantized data;
An image processing apparatus comprising : an information embedding unit that embeds the additional information in the second quantized data. A decoding unit which generates the first quantum Cadet over data by decoding the compressed data,
The image processing apparatus according to claim 1, further comprising: an encoding unit that encodes the second quantized data in which additional information is embedded by the information embedding unit to generate compressed data. The information embedding unit embeds the value after the embedding image processing device of the preceding than the value embedding the additional information to be closer to "0" according to claim 1 or 2 wherein. The image processing apparatus according to claim 1, further comprising: a table control unit that controls a value of the second quantization table based on a value of the first quantization table. The additional information belonging to a part is embedded element, the image processing apparatus according to claim 1 which is defined along the path of the zigzag scanning as a base point a DC component 4. And inverse quantization step of generating a first inverse quantizing the quantization cade over data frequency image data quantized using a first quantization table,
The frequency image data, the quantization step said quantized using different second quantization table and the first quantization table, thereby generating a second quantization data,
An information embedding step of embedding additional information in the second quantized data,
In the second quantization table element, at least the value of the element belonging to the portion in which the additional information is embedded is greater than the value of the element belonging to the portion in which the additional information is embedded in the first quantization table element. Even a small image processing method. And inverse quantization step of generating an inverse-quantized frequency image data using the first said quantum Cadet chromatography data of the first quantization table quantized using a first quantization table,
The frequency image data, the quantization step said quantized using different second quantization table and the first quantization table, thereby generating a second quantization data,
An information processing program comprising an information embedding step of embedding additional information in the second quantized data,
In the second quantization table element, at least the value of the element belonging to the portion in which the additional information is embedded is greater than the value of the element belonging to the portion in which the additional information is embedded in the first quantization table element. A recording medium on which a small image processing program is recorded so as to be readable by a computer.

Images