JP4011580B2 - Image data transmission system - Google Patents

Description

  The present invention relates to a data compression technique, and particularly to a compression technique suitable for transmission of image data.

  As a method for expressing digital data, a method of separately expressing a mantissa part and an exponent part is known. For example, Patent Document 1 discloses an apparatus that reduces storage capacity by expressing music data in a mantissa part and an exponent part. According to this apparatus, it is possible to obtain an effect that a sufficient dynamic range can be secured while the number of bits is small.

Japanese Patent Publication No. 1-445078

  However, Patent Document 1 does not consider a format suitable for data transmission. Therefore, when applied to data transmission, quality and speed are not always compatible.

  An object of the present invention is to solve the above problems and provide a compression technique particularly suitable for transmission of image data.

(1) (2) (3) An image data transmission system according to the present invention is an image data transmission system comprising a transmission device and a reception device,
The transmission device includes a data representation changing unit that divides given image data into mantissa part image data and exponent part image data, and a mantissa part compression unit that compresses the mantissa part image data to generate mantissa part compressed image data. And exponent part compression means for compressing exponent part image data to generate exponent part compressed image data, and transmission means for transmitting mantissa part compressed image data and exponent part compressed image data,
The receiving device includes: a receiving unit that receives mantissa compressed image data and exponent compressed image data; a mantissa decoding unit that generates mantissa image data by decoding the mantissa compressed image data; and exponent compressed image data And an exponent part decoding means for generating exponent part image data and display means for restoring and displaying the image data based on the mantissa part image data and the exponent part image data.

  The compression ratio can be increased by compressing the mantissa part and the exponent part separately.

(4) In the system according to the present invention, the mantissa compression means predicts the gradation of the target pixel based on the exponent image data of the pixels around the target pixel, and the predicted gradation and the actual gradation are obtained. Based on this, encoding is performed and compression is performed.

  Therefore, based on the predicted gradation, a short code is assigned to a gradation with high probability, and the compression rate can be increased.

(6) In the system according to the present invention, the transmitting unit of the transmitting device transmits the exponent part compressed image data with priority over the mantissa compressed image data, and the display unit of the receiving device receives only the exponent part image data. The pixel is displayed based on the exponent part image data, and when the mantissa part image data is received, the display of the pixel is corrected based on the mantissa part image data.

  Therefore, it is possible to perform image transmission with progressive image quality improvement.

(7) An image data transmission method according to the present invention is an image data transmission method comprising a transmission device and a reception device,
The transmitting device divides the given image data into mantissa part image data and exponent part image data, compresses the mantissa part image data to generate mantissa part compressed image data, compresses the exponent part image data, and compresses the exponent part image data Partial compressed image data is generated, exponential compressed image data is transmitted with priority over mantissa compressed image data,
The receiving device displays pixels that have received only exponent image compression data based on the exponent image data, and receives the mantissa image compression data and displays the pixels based on the mantissa image data. By correcting the image data, the image data is restored and displayed based on the mantissa image data and the exponent image data.

  The compression ratio can be increased by compressing the mantissa part and the exponent part separately.

  “Gradation conversion means” refers to means for converting numerical values representing gradation, color difference, color, and the like of given image data. In the embodiment, step S2 in FIG. 5, step S54 in FIG. 18, and step S62 in FIG. 19 correspond to this.

  “Mantissa exponent separation means” corresponds to step S3 in FIG. 5 in the embodiment.

  “Mantissa part compression means” refers to a unit that performs some kind of compression processing on mantissa part data. This concept includes not only lossless compression such as entropy compression but also lossy compression. In the embodiment, step S10 in FIG. 6 corresponds to this.

  The “exponential part compression means” refers to a unit that performs some kind of compression processing on the exponent part data. This concept includes not only lossless compression such as entropy compression but also lossy compression. In the embodiment, step S4 in FIG. 5 corresponds to this.

  “Transmission means” refers to a device that transmits data regardless of wireless or wired. In the embodiment, the transmission circuit 34 corresponds to this.

  "Receiving means" refers to a device that receives data regardless of whether it is wireless or wired. In the embodiment, the receiving circuit 54 corresponds to this.

  “Mantissa part decoding means” refers to a means for performing some kind of decoding processing on the compressed mantissa data. The concept includes not only decoding for lossless compression such as entropy compression but also decoding for lossy compression. In the embodiment, step S60 in FIG. 19 corresponds to this.

  The “exponential part decoding means” refers to something that performs some kind of decoding processing on the compressed exponent part data. The concept includes not only decoding for lossless compression such as entropy compression but also decoding for lossy compression. In the embodiment, step S52 in FIG. 18 corresponds to this.

  In the embodiment, the “display unit” corresponds to the image display unit 62.

BEST MODE FOR CARRYING OUT THE INVENTION

1. Overall Configuration FIG. 1 shows an overall configuration of an image data transmission system according to an embodiment of the present invention. The transmission device 14 includes gradation changing means 2, mantissa / exponent separation means 4, mantissa compression means 8, exponent compression means 10, and transmission means 12.

  Upon receiving the image data (0 to 255), the gradation converting means 2 converts the gradation expression into positive and negative data with the intermediate gradation (128) set to “0”. That is, as shown in FIG. 2A, image data indicated by gradations “0” to “255” is converted into converted image data indicated by gradations “-127” to “128” as shown in FIG. 2B. Convert to

  The mantissa / exponent separation means 4 receives the converted image data and separates it into exponent part image data and mantissa part image data. In this embodiment, the data expression changing means 6 is constituted by the gradation converting means 2 and the mantissa / exponent separating means 4.

  The mantissa part image data is compressed by the mantissa part compression means 8 and converted into mantissa part compressed image data. The exponent part image data is compressed by the exponent part compression means 10 to be exponent part compressed image data.

  The transmission means 10 transmits the exponent part compressed image data and the mantissa part compressed image data. At this time, the transmission means 10 transmits the exponent part compressed image data with priority over the mantissa part compressed image data. This enables a progressive process in which a rough gray image is transmitted first and a detailed gray image is transmitted later.

  The receiving device 16 includes a receiving unit 18, a mantissa part decoding unit 20, an exponent part decoding unit 22, gradation conversion units 24 and 26, and a display unit 28.

  The receiving means 18 receives the exponent part compressed image data and the mantissa part compressed image data transmitted from the transmission device 14. The mantissa part decoding unit 20 decodes the mantissa-compressed image data to form mantissa-part image data. The mantissa part image data is given to the gradation converting means 24, and is converted into the mantissa part image data having the gradation as shown in FIG. 2A by performing a conversion reverse to the gradation conversion performed on the transmission side.

  The exponent part decoding means 22 decodes the exponent part compressed image data to produce exponent part image data. The exponent part image data is given to the gradation conversion means 24, and is converted into the exponent part image data having the gradation as shown in FIG. 2A by performing a conversion reverse to the gradation conversion performed on the transmission side.

  The display means 28 reproduces the image data and displays the image based on the exponent part image data and the mantissa part image data.

When the image data indicated by the gradations “0” to “255” is divided into the exponent part image data and the mantissa part image data, the precision of the data having a low gradation (on the “0” side) is high. The accuracy decreases as the data becomes higher (“255” side). Therefore, in this embodiment, gradation conversion as shown in FIG. 2B is performed. As a result, it is possible to increase the accuracy in the vicinity of the intermediate gradation at which the human eye can best detect the difference in gradation.

2. Hardware Configuration of Transmission Device FIG. 3 shows a hardware configuration of the transmission device 16. A memory 30, a transmission circuit 34, and a hard disk 36 are connected to the CPU 32. For example, image data (still image) given from an external device or image data (still image) created by the CPU 32 itself is recorded in the memory 30. Data of each pixel is specified as coordinates in the x direction and the y direction.

  An operating system (such as TRON (trademark)) 38 and a control program 40 are recorded on the hard disk 36. The control program 40 performs its function in cooperation with the operating system 38.

In this embodiment, a program and image data are recorded on the hard disk 36. However, you may record on other recording devices, such as ROM and RAM.

3. Processing of Transmitting Device FIG. 5 shows a flowchart of the control program 40. The CPU 32 reads image data from the memory 30 (step S1). As shown in FIG. 7, the image data is, for example, 16 blocks × 16 pixels as one block. Therefore, the entire image data is composed of several blocks vertically and several blocks horizontally (see FIG. 8).

  In step S1, only image data for one pixel is read out. However, as described later, when processing for one pixel is completed, the process returns to step S1 to read image data of the next pixel. The order of pixel readout is the same as the normal image scanning direction.

  For example, assume that image data as shown in FIG. 11A is recorded in the memory 30. When the pixels in the x direction are sequentially read and processed from the upper left pixel and processed in the x direction, the next pixel in the y direction is read out from the left to the right and processed. This process is repeated until the pixel on the lower right is read and processed. In FIG. 11A, the image data is expressed in hexadecimal.

  Next, the read image data is subjected to gradation conversion by subtracting 127 to obtain converted image data (step S2). That is, the data system shown in FIG. 2A is converted to the data system shown in FIG. 2B. Thereby, data as shown in “Negative expression” in FIG. 9 is obtained corresponding to the original image data. In FIG. 9, the original image data is indicated by “binary value”.

  The converted image data obtained as a result of this conversion is as shown in FIG. 11B. In FIG. 11B, it is assumed that processing is performed for pixels with y = 2 and x = 2.

  Next, the CPU 32 separates the converted image data into exponent part image data and mantissa part image data (step S3). As a result, data as shown in “sign bit”, “mantissa part”, and “exponent part” in FIG. 9 is obtained. Since the mantissa part is 4 bits, a slight error occurs as shown in “Reproduced binary value”.

  The CPU 32 records the mantissa part image data and the exponent part image data thus obtained in the memory 30 in association with each other as shown in FIG. At this time, coordinates “y, x” specifying the pixel are also recorded.

  A schematic representation of the mantissa image data and the exponent image data is shown in FIG. 11C. In this figure, it is shown in binary number. The upper part is the exponent part and the lower part is the mantissa part.

  Next, the CPU 32 compresses the exponent image data (step S4). In this embodiment, compression is performed by entropy coding. As shown in FIG. 12, the CPU 32 displays the upper left, right above, upper right, and left pixels P (1,1), P (1,2), P (1,3) of the target pixel P (2,2), The exponent image data is acquired for P (2,1). That is, the exponent image data of adjacent pixels that have already been processed is read from the memory 30.

  Subsequently, the CPU 32 predicts exponent part pixel data of the target pixel P (2, 2) based on the readout exponent part image data of surrounding pixels. In this embodiment, the prediction is performed in consideration of the high probability of taking exponent part pixel data similar to the exponent part image data of adjacent pixels.

  Specifically, as shown in FIG. 13, a normal distribution curve α is assumed with the predicted exponent pixel data as e ′ and the actual exponent pixel data as e. 13, the probability that “e ′” is equal to “e” is 0.5265, and the probability that “e” is equal to “e′−1” is 0.125... That is, the CPU 32 records the probability as shown in FIG. 14A. Note that this probability may be prepared in advance as a table for each pattern of surrounding exponent image data.

  Next, the CPU 32 determines a code to be assigned to each numerical value based on the calculated appearance probability. In this embodiment, a Huffman tree is generated based on the probability, and a Huffman code for each numerical value is assigned as shown in FIG. 14B.

  For example, in the above example, “1” is assigned to “e ′”, “011” is assigned to “e ′ + 1”, “010” is assigned to “e′−1”, and so on. Create a table. In this way, a short code is assigned to data with a high appearance probability.

  Next, the CPU 32 encodes the exponent image data of the target pixel P (2, 2) based on the created code table. In FIG. 12A, if the predicted exponent image data e ′ of the target pixel P (2, 2) is “+01”, it is equal to “+01” of the actual exponent image data e ′, so e ′ = e. The code “1” is assigned from FIG. 14B. The CPU 32 records the exponent part compressed image data calculated in this way in the memory 30. FIG. 15 shows the recorded index portion compressed image data.

  Next, the CPU 32 determines whether or not the exponent compressed image data has been calculated for all the pixels in the target block (step S5). If an unprocessed pixel remains, step S1 is repeatedly executed for the next pixel. If the exponent compressed image data is calculated for all the pixels, the process proceeds to step S6.

  In step S6, the CPU 32 reads out the exponent part compressed image data of the target block recorded in the memory 30 and gives it to the transmission circuit 34 (step S6). The transmission circuit 34 records the exponent compressed image data given from the CPU 32 in a buffer and sequentially transmits the data.

  Next, the CPU 32 determines whether or not all blocks have been processed (step S7). If the remaining blocks remain, the process after step S1 is repeated with the next block as the target block.

  When all the blocks are processed and the exponent compressed image data is given to the transmission circuit 34, the CPU 32 determines whether there is a margin in the transmission path (step S8). The transmission circuit 32 sequentially transmits the data stored in the buffer. Therefore, by knowing the amount of untransmitted data stored in the buffer, it is possible to indirectly know the communication speed of the current transmission path. This is because the amount of data stored in the buffer is small when the communication speed (communication band) is high, and the amount of data stored in the buffer is large when the communication speed (communication band) is low. The transmission circuit 32 notifies the CPU 32 of the amount of untransmitted data stored in the buffer. Thereby, the CPU 32 can know the current communication speed (communication band).

  If there is no room in the bandwidth of the transmission path, the CPU 32 proceeds to processing of the next image. That is, for the next image data (still image), the processing from step S1 is executed.

  If there is room in the bandwidth of the transmission path, the CPU 32 performs transmission processing of mantissa image data. First, the CPU 32 reads out the mantissa image data of the first pixel of the first block from the memory 30. The mantissa image data is generated in step S3 and recorded in the memory 30 as shown in FIG. The reading order of the mantissa image data is also the same as the reading order of the exponent part image data. That is, one pixel is read out zigzag from the upper left pixel toward the lower right pixel.

  For example, as shown in FIG. 12B, it is assumed that the mantissa image data of the pixel P (2, 2) has been read. The CPU 32 extracts only the upper 2 bits of the read mantissa image data and performs compression by entropy coding. Similar to the exponent image data, the prediction is performed to determine the code. At this stage, since the values of the exponent image data of the surrounding 8 pixels are known, prediction is also performed using these values.

  For example, as shown in FIG. 12B, when the target pixel is P (2, 2), the surrounding pixels P (1, 1), P (1, 2), P (1, 3), P ( 2,1), P (2,3), P (3,1), P (3,2), and P (3,3).

  At this time, both the exponent part image data and the mantissa part image data are known for the pixels P (1,1), P (1,2), P (1,3), and P (2,1). Based on these, image data is restored and used for prediction. Since only the exponent part image data is known for the pixels P (2,3), P (3,1), P (3,2), and P (3,3), the mantissa part image data is used. The median value is adopted to restore the image data, and this is used for prediction.

  The CPU 32 uses the surrounding pixels P (1,1), P (1,2), P (1,3), P (2,1), P (2,3), P (3,1), P ( 3,2) and image data of P (3,3) are predicted based on the image data of the target pixel P (2,2). The prediction method is the same as the prediction of exponent image data.

  The point of assigning the Huffman code based on the probability of the difference between the predicted image data e 'and the actual image data e is the same as the encoding of the exponent image data. In this way, the mantissa image data is encoded.

  Next, the CPU 32 determines whether or not processing has been performed for all pixels of the target block (step S11). If the mantissa-compressed pixel data is generated for all the pixels of the target block, the mantissa-compressed pixel data recorded in the memory 30 is read and supplied to the transmission circuit 34. Thereby, the transmission circuit 34 transmits the mantissa-compressed pixel data.

  Next, the CPU 32 determines whether or not all blocks have been processed (step S13). If an unprocessed block remains, step S8 and subsequent steps are executed with the next block as the target block.

  When there is no room in the transmission path or when the processing for all the blocks is completed, the CPU 32 proceeds to processing of the next image data. As described above, the image data is compressed and transmitted.

  In the above embodiment, in step S9, the upper 2 bits are extracted and the lower bits are not targeted for transmission. However, after the upper 2 bits of the mantissa-compressed image data is transmitted for all the pixels, the lower-bit mantissa-compressed image data may be transmitted (see the second transmission in FIG. 16).

  Further, an error generated when the exponent part and the mantissa part are separated may be recorded in the memory 30 and transmitted after the lower bits are transmitted.

Further, in the separation into the exponent part and the mantissa part shown in FIG. 9, regarding the Q and R parts in the figure, the exponent part is not “00” and the first bit of the mantissa part is “1”. Can be distinguished from others in terms of points. Therefore, the mantissa part may be configured as shown in FIG. 17 by omitting the first bit “1” of the mantissa part. As a result, it is possible to improve the accuracy of the Q and R portions. On the receiving side, when the exponent part is not “00”, the original number can be reproduced by adding “1” as the first bit of the mantissa part.

4). Hardware Configuration of Receiving Device FIG. 4 shows a hardware configuration of the receiving device. To the CPU 52, a buffer 50, a receiving circuit 54, a ROM 56, an image display unit 62, and a frame memory 64 are connected. The reception circuit 54 receives data transmitted from the transmission device 16 by wire or wireless. The image display unit 62 displays an image by processing of the CPU 52.

An operating system (such as TRON (trademark)) 58 and a control program 60 are recorded in the ROM 56.

5). Processing of Receiving Device FIGS. 18 and 19 are flowcharts of the control program 60. The receiving circuit 54 stores the received exponent part compressed image data and mantissa part compressed image data in the buffer 50. The CPU 52 reads the exponent compressed image data for the target pixel of the target block recorded in the buffer 50. The pixel readout order is the same as that in the transmission apparatus.

  Next, the CPU 52 decodes (expands) the exponent part compressed image data. For example, as shown in FIG. 12A, when processing the pixel P (2, 2), the target pixel is based on the exponent image data of the surrounding pixels in the same manner as the processing performed in the transmission device 16. The numerical value of is predicted. Let the predicted exponent image data be e '.

  Next, the CPU 52 creates the same code table (FIG. 14B) as the one created by the transmission device 16 and records it in the buffer 50. Then, referring to this table wp, the exponent image data is restored based on the predicted exponent image data e 'and the exponent compressed image data (Huffman code). The restored exponent image data is recorded in the buffer 50.

  Next, the CPU 52 restores the exponent image data to converted image data (step S53). Although there is no mantissa part image data, if there is exponent part image data, an approximate value of the converted image data can be obtained.

  Further, the CPU 52 obtains image data by shifting the gradation of the converted image data (step S54). That is, “127” is added to the data value. Thereby, the data system as shown in FIG. 2B is converted into the data system as shown in FIG. 2A. The image data obtained by the conversion is recorded in the buffer 50.

  Next, the CPU 52 determines whether image data has been obtained for all the pixels of the target block (step S55). If there is an unprocessed pixel, step S51 and subsequent steps are repeatedly executed with the next pixel as the target pixel. When the process is completed for all the pixels of the target block, the image data of the target block recorded in the buffer 50 is read out and transferred to the corresponding position in the frame memory 64 (step S56). As a result, the image of the block is displayed on the image display unit 62.

  Next, the CPU 52 determines whether or not all blocks have been processed (step S57). If there is an unprocessed block, step S51 is repeatedly executed with the next block as the target block. When processing is completed for all blocks, the entire image is displayed on the image display unit 62. Since there is no data of the mantissa part, it is possible to express the approximate brightness although the accuracy is low.

  When the processing is completed for all the blocks, the CPU 52 determines whether or not the mantissa-compressed image data for the image is recorded in the buffer 50 (step S58). If not recorded, the process proceeds to the next image data. That is, step S51 and subsequent steps are executed for the next image.

  If the mantissa-compressed image data exists, the mantissa-compressed image data of the target pixel of the target block is read (step S59). The pixel readout order is the same as that in the transmission apparatus.

  The CPU 32 decompresses the mantissa compressed image data and restores the mantissa image data. This decompression process can be realized by executing the compression process in the transmission apparatus in reverse.

  For example, as shown in FIG. 12B, when the pixel P (2, 2) is being processed, the numerical value of the target pixel is determined based on the data of the surrounding pixels in the same manner as the processing performed in the transmission device 16. Make a prediction. Here, it is assumed that the predicted image data is e ′.

  Next, the CPU 52 creates the same code table as that created by the transmission device 16 and records it in the buffer 50. Then, referring to this table, the mantissa image data is restored based on the predicted image data e ′, the exponent compressed image data (Huffman code), and the mantissa compressed image data (Huffman code) (step S60). ). The restored mantissa image data is recorded in the buffer 50.

  The CPU 52 restores the converted image data based on the mantissa image data obtained by decoding and the exponent image data of the pixel. Thereby, converted image data (image quality improvement) can be obtained (step S61). Further, the CPU 52 performs gradation conversion on the converted image data (image quality improvement) to restore the image data for improving the image quality (step S62). The image data for improving the image quality is recorded in the buffer 50.

  Next, the CPU 52 determines whether or not image data for improving image quality has been obtained for all pixels of the target block (step S63). If there is an unprocessed pixel, step S59 and subsequent steps are repeatedly executed with the next pixel as the target pixel. When the processing is completed for all the pixels of the target block, the image data for improving the image quality of the target block recorded in the buffer 50 is read out and added to the image data based on the exponent image data. Transfer to the corresponding position (step S64). As a result, an image with improved image quality of the block is displayed on the image display unit 62.

  Next, the CPU 52 determines whether or not all the blocks have been processed (step S65). If there is an unprocessed block, step S59 is repeatedly executed with the next block as the target block. When the processing is completed for all the blocks, the entire image with improved image quality is displayed on the image display unit 62. The CPU 52 proceeds to processing of the next image data. In this way, progressive display can be realized.

When the mantissa data is sent from the transmission device 16 in several times, image data for improving image quality may be generated and displayed sequentially.

6). Other Embodiments In the above embodiment, the case where a still image is transmitted has been described. However, a moving image may be transmitted similarly. Further, a numerical value of the α plane representing the transparency may be a transmission target.

  Further, when transmitting a moving image or the like, the image data may not be a transmission target, but a difference image from the previous frame may be the transmission target. Since the difference image has positive and negative numerical values centered on “0”, gradation conversion is not necessary.

  When transmitting a moving image, motion prediction difference image data obtained as a result of motion compensation may be the transmission target. Similar to the above embodiment, it is possible to perform prediction and encoding based on surrounding prediction difference image data.

  In both cases of still images and moving images, encoding may be performed after orthogonal transform (DCT or the like) for each block. In this case, encoding can be performed by performing prediction based on the value of the same frequency component of the adjacent block.

  In the embodiment described above, image data is targeted, but other data such as audio data can also be targeted.

  In the above embodiment, a monochrome image has been described. Similarly, the present invention can be applied to a color image.

  In the above embodiment, each function of FIG. 1 is realized by software, but a part or all of the functions may be configured by hardware logic.

1 is an overall configuration diagram of an image data transmission system according to an embodiment of the present invention. It is a figure which shows the gradation conversion of image data. It is a figure which shows the structure of a transmitter. It is a figure which shows the structure of a receiver. It is a flowchart of a control program. It is a flowchart of a control program. It is a figure which shows a pixel and a block. It is a figure which shows a block structure It is a figure which shows the correspondence of image data, mantissa part data, and exponent part data. It is a figure which shows the mantissa part data and exponent part data which were recorded on the memory. It is a figure which shows the correspondence of image data, mantissa part data, and exponent part data. It is a figure for demonstrating a prediction process. It is a figure which shows calculation of the probability based on normal part distribution. It is a figure for demonstrating Huffman encoding. It is a figure which shows exponent part compression image data. It is a figure which shows transmission of error data. It is a figure which shows the correspondence of image data, mantissa part data, and exponent part data. It is a flowchart of the control program of a receiver. It is a flowchart of the control program of a receiver.

Explanation of symbols

2 ... gradation conversion means 4 ... mantissa / exponent separation means 8 ... mantissa compression means 10 ... exponent compression means 12 ... transmission means

Claims (2)

An image data transmission system comprising a transmission device and a reception device,
The transmitter is
Data representation changing means for dividing the given image data into mantissa image data and exponent image data;
Mantissa compression means for compressing mantissa image data to generate mantissa compression image data;
Exponent part compression means for compressing exponent part image data to generate exponent part compressed image data;
Transmitting means for transmitting the mantissa compressed image data and the exponent compressed image data, the transmitting means for transmitting the exponent compressed image data with priority over the mantissa compressed image data;
With
The receiving device is:
Receiving means for receiving mantissa compressed image data and exponent compressed image data;
A mantissa decoding means for decoding the mantissa compressed image data to generate mantissa image data;
Exponent part decoding means for decoding exponent part compressed image data to generate exponent part image data;
Display means for restoring and displaying the image data based on the mantissa image data and the exponent image data, and for pixels that have received only the exponent image compressed data, display based on the exponent image data, Upon receiving the mantissa image compressed data, display means for correcting the display of the pixel based on the mantissa image data;
An image data transmission system comprising: An image data transmission method comprising a transmission device and a reception device,
The transmitter is
The given image data is divided into mantissa image data and exponent image data,
Compress the mantissa image data to generate mantissa compressed image data,
Compress the exponent image data to generate exponent compressed image data,
Send exponential compressed image data with priority over mantissa compressed image data,
The receiving device is:
Receive mantissa compressed image data and exponent compressed image data,
Decoding the mantissa-compressed image data to generate mantissa-part image data;
Decoding exponential compressed image data to generate exponential image data;
For pixels that have received only the exponent image compression data, display based on the exponent image data, and upon receiving the mantissa image compression data, by correcting the display of the pixel based on the mantissa image data, The image data is restored and displayed based on the mantissa image data and the exponent image data,
An image data transmission method characterized by the above.

Images