JP2008172546A - Image encoding/decoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that, while it can be considered to perform a search by 4-pixel accuracy (3-pixel thinning) horizontally since a wider region should be searched in a motion search because an image size is considerably bigger and an aspect ratio is higher in an HDTV compared with a conventional SDTV though there is a method of searching motions after thinning every other pixel horizontally in the SDTV, image quality substantially is deteriorated since a pixel pitch is coarse originally in the SDTV when encoding is performed keeping the thinning rate during the encoding of the SDTV. <P>SOLUTION: A motion search range and a pixel thinning rate are adaptively switched according to the image size. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image encoding / decoding device.

  As background art in this field, for example, there is JP-A-2004-349756. According to the publication, “[Problem] Realizing moving image coding that performs a search with high motion search accuracy and a small motion compensation unit for input video at low cost. [Solution] Input by image enlargement unit 10 The video is expanded M times in the horizontal direction and N times in the vertical direction, and the resulting enlarged image is encoded by the preprocessing encoding unit 11. The main encoder 13 is obtained by the preprocessing encoding unit 11. Receiving the motion vector information, the size of the motion vector is reduced to 1 / M in the horizontal direction and 1 / N in the vertical direction, and motion compensation is performed using the vector obtained for the input video, Encoding "(see summary).

  As background art in this field, for example, there is JP-A-2002-135784. In this publication, “[Problem] To provide a video coding method using motion estimation and compensation with an adaptive motion search range. [Solution] A fixed motion search range is used for very fast motion. Sometimes a true match area cannot be found, and a fixed motion search range wastes a lot of processing time for very slow motions. “The use of the reduction method increases the efficiency of motion estimation, and as a result, improves the encoding efficiency.” (See summary).

JP 2004-349756 A Japanese Patent Laid-Open No. 2002-135784

  In image coding, motion search is a very heavy process, and hardware usually performs a search with a narrow search range. One of them is a technique called hierarchical search. In the hierarchical search, a rough search is first performed, and the surroundings are further finely searched for the result.

  Here, the rough search first performed in the hierarchical search is an important process that greatly affects the image quality. If the motion search range can be made as large as possible, the possibility that similar images can be found in a wider range is increased, and the image quality is improved. However, if the motion search range is unnecessarily large, the time required for the search increases in proportion to it, and the processing time increases. Thus, as an example, in SDTV, there is a method of performing a search after thinning out every other pixel horizontally.

  However, the HDTV has a considerably larger image size than the conventional SDTV, and has a wide aspect ratio, so a wider area must be searched for by motion search. Therefore, as an example, in HDTV, it is conceivable to search horizontally with 4-pixel accuracy (3 pixel thinning).

  However, when SDTV encoding is performed with this thinning rate, SDTV has a rough pixel pitch, resulting in a significant decrease in image quality.

  Therefore, in the present invention, as an example, the motion search range and the pixel thinning rate are adaptively switched according to the image size.

  According to the present invention, as an example, an effect of improving the image quality can be obtained.

  Problems, configurations, and effects other than those described above will be described in the following embodiments.

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

  FIG. 1 shows a block diagram of an image encoding / decoding device 1 which is an embodiment of the present invention. First, the operation during encoding will be described. A moving image is input from the image input terminal and received by the video I / F unit 12. The received moving image is temporarily stored in a buffer area provided in the SDRAM 2 via the SDRAM I / F unit 17. After a predetermined delay time, the moving image data is read from the buffer area in the SDRAM 2 and sent to the encoder unit 13. Encoder unit 13 performs an encoding process. The encoded data (elementary stream) after the encoding process is sent to the Mux / Demux unit 15. The System Mux / Demux unit 15 performs system stream processing such as header code, time stamp addition, and packetization. The encoded data that has been converted into the system stream is output as an encoded image to an external device via the Device I / F unit 16.

  Next, the operation at the time of decoding will be described. First, encoded image data is input to the Device I / F unit 16 from an external device. The encoded image data received by the Device I / F unit 16 is input to the System Mux / Demux unit 15. The System Mux / Demux unit 15 removes a header code from the input encoded image data and extracts an elementary stream. The extracted elementary stream is sent to the decoder unit 14. In the decoder unit 14, the elementary stream is decoded. The moving image data that has been decrypted is sent to the Video I / F unit 12. The moving image data sent to the Video I / F unit 12 is stored in a buffer area provided in the SDRAM 2 through a decoding process. After a predetermined delay time, the moving image data is read from the buffer area in the SDRAM 2 and output from the Video I / F unit 12 to the outside of the apparatus 1 as an image output.

  Further, as shown in FIG. 4, transcoding can be performed by using the Decoder unit 14 and the Encoder unit 13. In this case, encoded image data is input to the Decoder unit 14, decoded as described above, and moving image data is output. The moving image data is input to the Encoder unit 13, encoded as described above, and encoded image data is output. At this time, as shown in FIG. 4, by exchanging information regarding encoding between the processor of the decoder unit 14 and the processor of the encoder unit 13, the encoder unit 13 can perform more efficient encoding.

  FIG. 2 shows a block diagram of the encoder unit 13. It consists of a main processor 131 and a sub processor 132, and processing blocks 133 to 1310 controlled by these processors. The input image is input to the in-screen prediction unit 133 and the motion search unit 134. The intra-screen prediction unit 133 creates a predicted image of the partial image currently being processed from the already encoded portion in the screen. In addition, the motion search unit 134 searches for an image closer to the partial image currently being processed from other previously encoded images, and creates a predicted image. The more advantageous one of the predicted images output from the in-screen prediction unit 133 or the motion search unit 134 is selected. This selection decision is made with the main processor 131 involved. One block finally selected obtains the difference between the predicted image and the current image. The prediction error as a result is output to the DCT / quantization processing unit 135. The finally selected block outputs a prediction image to the motion compensation unit 137.

  The DCT / quantization processing unit 135 performs discrete cosine transform or a predetermined frequency transform process equivalent to the input prediction error. Subsequently, a quantization process using a quantization coefficient is performed. The processed result is output to the entropy encoding unit 139.

  The entropy encoding unit 139 performs a predetermined entropy encoding process based on an instruction from the sub processor 132. The processing result is output via the buffer unit 1310.

  The output of the DCT / quantization processing unit 135 is input to the inverse quantization / IDCT processing unit 136. The inverse quantization / IDCT processing unit 136 performs inverse quantization processing and inverse discrete cosine transform (or predetermined equivalent processing), obtains a prediction error, and outputs the result to the compensation unit 137.

  The motion compensation unit 137 adds the input prediction error and the predicted image to create a restored image. The restored image is output to the deblocking unit 138.

  The restored image input to the deblocking unit 138 is subjected to block boundary filtering. The local decoded image as a result is used in the motion search unit 134 in the future as a motion search image.

  The processing of the motion search unit 134 will be described with reference to FIG. Image 3 is an image that has been encoded in the past, and image 4 is an image that is currently being encoded. The macro block 41 is a partial image currently being encoded, and the macro block 31 is a macro block corresponding to the same position as the macro block 41 in the image 3. A region 32 having a width H and a height V around the macroblock 31 is taken as a motion search range.

  FIG. 6 shows the motion search range 32. The motion search unit 134 performs processing for finding a portion closest to the macroblock 41 within the motion search range 32. Here, if the ranges H and V can be made as large as possible, the possibility that similar images can be found in a wider range becomes higher. However, if H and V are unnecessarily large, the time required for the search increases in proportion thereto, and the processing time increases. Therefore, the following measures are taken.

That is, an image is prepared by thinning out pixels of the search range 32 and the macroblock 41 in units of 2 pixels in the horizontal direction. In this way, the number of pixels to be processed is H / 2 in the horizontal direction, and the processing time can be halved. Alternatively, if an image obtained by thinning out the pixels of the search range 32 and the macroblock 41 in units of 4 pixels in the horizontal direction is prepared, the processing time is similarly reduced to a quarter.
Alternatively, if the processing time is the same, it is possible to search a search range of 2 times or 4 times, respectively, and the image quality is improved. In particular, when encoding a high-definition television signal having a large number of pixels, it is necessary to ensure a wide search range. Therefore, a wide search range can be secured by selecting a unit of four pixels. In the case of a conventional television signal, it is possible to use thinning-out in units of two pixels by prioritizing a fine search over a wide search range.

As described above, according to the present invention, it is possible to adaptively select the search range decimation rate and the size of the search range in accordance with the image size.
In this manner, the motion search unit 134 searches for the closest search to the current image macroblock 41 in the motion search range 32.

  FIG. 7 is a diagram for explaining the continued operation of the motion search unit 134. The pixel position found above is further searched in a range of horizontal ± 2 pixels and vertical ± 2 pixels. A region A of 8 × 8 pixels in FIG. 7 is searched using a PE (processor element) group A capable of simultaneously performing a search for a total of 25 pixels including the central pixel. Similarly, the search is executed in parallel on PE-B, C, and D for adjacent B × 8 and D × 8 pixel regions B, C, and D.

  In this way, the nearest image is searched for each of the areas PE-A to D, and as a result, the respective position offsets, that is, motion vectors are obtained.

  Further, the motion search unit 134 searches the horizontal and vertical regions of ± 0.5 in the vicinity of the motion vector in increments of 0.25 pixels. The above-described PE-A to D are used again for this search.

  Next, the decoder unit 14 will be described with reference to FIG. It comprises a main processor 141 and sub-processor 142, and processing blocks 143 and 146 to 1410 controlled by these processors.

  The input elementary stream is buffered by the buffer unit 1410 and then input to the entropy decoding unit 149. As shown in FIG. 8, the entropy decoding unit 149 performs a predetermined entropy decoding process based on an instruction from the sub processor 142. The macroblock layer parameters and data, which are the processing results, are input to the inverse quantization / IDCT unit 146. Further, the upper layer such as a sequence or a picture header is analyzed by the sub processor 142 and output as upper layer information.

The inverse quantization / IDCT processing unit 146 performs inverse quantization processing and inverse discrete cosine transform (or predetermined equivalent processing), restores a prediction error, and outputs the result to the compensation unit 147.
The motion compensation unit 147 adds the input prediction error and the prediction image to create a restored image. Note that, as the predicted image here, either motion compensation or an intra-screen predicted image is used depending on the content of the code. In the case of the intra prediction image, the prediction image created by the intra prediction unit 143 is used. The restored image created by the motion compensation unit 147 is output to the deblocking unit 148.

  The restored image input to the deblocking unit 148 is subjected to block boundary filtering and output as an output image.

  Among the entropy codes that can be handled by the entropy decoding unit 149, arithmetic code decoding processing will be described with reference to FIG. In the figure, 1491 is a Bin decoder unit that performs arithmetic decoding processing to obtain a 1-bit value “Bin”, 1492 is a numerical value restoration unit that obtains a set of numerical values from a plurality of Bin columns, and 1493 is a Bin decoder unit It is a Context calculation part which calculates the parameter "Context value" used for the decoding process in.

  FIG. 10 illustrates the Context calculation unit 1494 of the present invention in more detail. A Context # 0 calculation unit 14941 and a Context # 1 calculation unit 14942 are incorporated therein. The Context # 0 calculation unit 14941 starts the Context calculation assuming that the result Bn-1 as the previous decoding result is 0 before the Bin value Bn-1 is finalized. Similarly, the Context # 1 calculation unit 14942 starts the Context calculation assuming that the result Bn-1 of the immediately preceding decoding result Bin is 1 before the value Bn-1 is determined. These results are input to the Bin decoder unit 1491 together as Context [0] and [1]. As soon as the previous decoding result Bn-1 is determined, the Bin decoder unit 1491 selects an appropriate one of the input Context [0] or [1] based on the value, and performs the next decoding. Use.

  As described above, according to the arithmetic decoding process of the present invention, since the Context operation can be started before the immediately preceding Bin decode value is determined, the processing time of the feedback loop inside the arithmetic decoding process can be shortened. It is possible to significantly increase the processing speed of the arithmetic decoding process.

1 is a block diagram of an image encoding / decoding device 1 according to an embodiment of the present invention. The block diagram of the Encoder part 13 is shown. The block diagram of the Decoder part 14 is shown. It is a figure which shows the Example in the case of transcoding. It is a figure explaining the process of the motion search part. A motion search range 32 is shown. FIG. 6 is a diagram for explaining the operation of a motion search unit 134. It is a figure explaining the entropy decoding part 149. FIG. It is a figure explaining the decoding process of an arithmetic code. It is a figure explaining the Context calculation part 1494. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Image encoding / decoding apparatus 2 ... SDRAM
12. ・ Video I / F
13 · Encoder unit 14 · Decoder unit 15 · System Mux / Demux unit 16 · Device I / F
17. · SDRAM I / F

Claims (4)

An image encoding / decoding device that performs motion search,
An input means for inputting image data;
First motion search means for performing a rough motion search on input image data;
Second motion search means for performing a motion search more finely than the first motion search means using the motion search result output from the first motion search means;
With
The accuracy of the motion search performed by the first motion search means is determined according to the image size of the input image data;
An image encoding / decoding device characterized by the above. In claim 1,
The first motion search means performs a motion search by thinning the input image data in the horizontal direction, and the thinning rate is determined according to the image size of the input image data.
An image encoding / decoding device characterized by the above. In claim 1 or 2,
The first motion search means performs a motion search by thinning input image data in the horizontal direction, and the larger the image size of the input image data, the larger the thinning rate,
An image encoding / decoding device characterized by the above. In any one of Claims 1 thru | or 3,
When the input image data is a standard television signal, the first motion search means performs a motion search by thinning out every other pixel in the horizontal direction, and when the input image data is a high-definition television signal. Performing motion search by thinning out every 3 pixels in the horizontal direction;
An image encoding / decoding device characterized by the above.

Images