JP2011091772A - Image encoder - Google Patents

Abstract

PROBLEM TO BE SOLVED: To adaptively switch an initial quantization parameter determination method according to a scene.
When a change amount of an image is equal to or greater than a predetermined threshold, a first method for determining an initial quantization parameter by predicting a generated code amount from information of an original image of an encoding target image is selected. Initial quantization for an encoding target image by selecting a second method for predicting a generated code amount of an encoding target image from an encoded image and determining an initial quantization parameter when less than a predetermined threshold Determine the parameters.
[Selection] Figure 1

Description

  The present invention relates to an image encoding device, and more particularly to a quantization control technique.

  MPEG (Moving Picture Experts Group) -2 is known as a high-efficiency encoding method for moving image data. In recent years, H.264 has been developed by ITU-T (International Telecommunication Union Telecommunication Standardization Sector) and ISO (International Organization for Standardization) as an encoding method that achieves higher compression than MPEG-2. H.264 / MPEG-4 part 10 has been standardized. H. In H.264 / MPEG-4 part 10 (hereinafter referred to as H.264), in addition to intra-prediction coding and inter-prediction coding, adaptive code amount control is performed, thereby preventing high image quality degradation. Is possible.

  MPEG-2 and H.264 As a code amount control method used in an encoding method such as H.264, the Test Model 5 (TM5) method is known. In the TM5 system, a quantization characteristic value calculated from an encoded image is set for each type of image to be encoded (picture type). Picture types are intra-picture coded pictures (I pictures), forward predictive coded pictures (P pictures), and bi-directional predictive coded pictures (B pictures). For example, when the encoding target picture is an I picture, the quantization characteristic value of the encoding target I picture is set based on the encoding result of the encoded I picture. In this specification, the determination of the quantization parameter by predicting the generated code amount of the encoding target image from the encoded image is referred to as a “feedback method”. Details of the TM5 method are disclosed in Patent Document 1.

  When determining the initial quantization parameter that is the quantization characteristic value of the encoding target image based on the prediction of the generated code amount using such a feedback method, the generated code amount is determined based on the past encoded image. In order to make a prediction, inconvenience occurs in the following situation. For example, this is a situation in which a picture has changed greatly from an encoded image, such as a scene with intense movement, or a scene change has been performed. In such a situation, the accuracy of prediction of the generated code amount is lowered, and the code amount cannot be accurately distributed, resulting in a problem that coding efficiency is lowered.

  On the other hand, Patent Document 2 is known as a proposal relating to a code amount control method different from the feedback method. In the method described in Patent Document 2, coarse motion vector calculation using a reduced image is performed before obtaining a detailed motion vector for an original image. Then, the degree of consistency between the reduced prediction block and the reference block is statistically analyzed, and the statistical analysis result is used for code amount allocation control. In this specification, determining the quantization parameter by predicting the generated code amount from the information of the original image of the encoding target image is referred to as “feedforward method” in this specification.

JP-A-10-285589 JP-A-11-298904

  With the above-mentioned code amount control method using the feed-forward method, the accuracy of code amount estimation can be improved in situations where there is intense motion or scene changes compared to when the initial quantization parameter is determined by the feedback method. Is possible.

  However, for scenes with little motion, the difference between images is small and the relevance is high, so it is known that the feedback method that determines the initial quantization parameter using encoded image data is more accurate. It has been.

  FIG. 4 shows a comparison result between the feedback method (FB method) and the feedforward method (FF method). As described above, the feedback method uses the encoded image data that is temporally close to the encoding target image to obtain the initial quantization parameter. When there is little relevance between images, the evaluation is high in terms of prediction accuracy and the like. On the other hand, since the feed-forward method performs code amount control for each image, a constant evaluation can be obtained regardless of scene changes. Because accuracy increases, evaluation is lower than feedback method.

  As described above, there is a problem that the initial quantization parameter determination method based on the feedback method and the initial quantization parameter determination method based on the feedforward method are disadvantageous depending on the encoding target image.

  Accordingly, an object of the present invention is to perform suitable quantization control by adaptively switching the initial quantization parameter determination method according to the scene.

  In order to achieve the above object, an image encoding apparatus according to the present invention is an image encoding apparatus that encodes a moving image, and a quantization unit that quantizes data of an encoding target image included in the moving image A first method for predicting a generated code amount from information on an original image of the encoding target image and determining an initial quantization parameter; and a generated code amount of the encoding target image from a coded image And a second control method for determining an initial quantization parameter selectively, a quantization control means for controlling the quantization in the quantization means according to the determined initial quantization parameter, the encoding target image, Calculating means for calculating an image change amount between a predetermined encoded image and the quantization control means when the image change amount calculated by the calculation means is greater than or equal to a predetermined threshold value. Select one method By image change amount calculated to select the second method in the case of less than the predetermined threshold value by said calculating means, and determining an initial quantization parameter for the encoding target image.

  According to the present invention, it is possible to perform suitable quantization control by adaptively switching the initial quantization parameter determination method according to the scene.

It is a block diagram which shows schematic structure of the image coding apparatus which concerns on Example 1 of this invention. It is a flowchart which shows the initial quantization parameter determination method. It is a block diagram which shows schematic structure of the image coding apparatus which concerns on Example 2 of this invention. It is a figure which shows the comparison result of a feedback system and a feedforward system.

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

  FIG. 1 is a block diagram illustrating a schematic configuration of an image encoding device 100 according to the first embodiment of the present invention. The image encoding device 100 is mounted on a digital video camera or the like and can be applied as an image encoding device that compresses and encodes captured image data. In addition, the image encoding apparatus 100 is an H.264 format. Configuration for performing adaptive code amount control (quantization control) in addition to intra prediction encoding (intra prediction encoding) and inter prediction encoding (inter prediction encoding) corresponding to the H.264 encoding scheme Is provided. The configuration of the image encoding device 100 will be described below with reference to the block diagram of FIG.

  The input image data input to the image encoding device 100 is a moving image captured mainly by a camera (not shown) or the like. The image encoding device 100 performs an encoding process for each screen (frame) constituting the moving image. The input image data is supplied to the difference calculation unit 101, the motion search unit 114, and the reduced image generation unit 112, respectively. The difference calculation unit 101 calculates a difference value between input image data to the image encoding device 100 and predicted image data generated by a predicted image generation unit 107 described later. The difference value is image residual data.

  Next, the orthogonal transform unit 102 receives the image residual data output from the difference calculation unit 101, performs orthogonal transform processing by discrete cosine transform (DCT) in units of pixel blocks, and outputs DCT coefficients. To do. DCT is performed in units of 8 × 8 pixels or 4 × 4 pixel blocks. The quantization unit 103 performs a quantization process based on the DCT coefficient output from the orthogonal transform unit 102 and the quantization parameter output from the quantization parameter control unit 115 described later. Next, the entropy encoding unit 104 performs binarization processing and arithmetic encoding processing on the quantized data output from the quantization unit 103, generates encoded data, and outputs the encoded data. Information on the amount of generated code in the entropy encoding unit 104 is supplied to a quantization parameter control unit 115 described later.

  On the other hand, the quantized data output from the quantization unit 103 is also output to the inverse quantization unit 105. This is used to generate predicted image data. The inverse quantization unit 105 performs inverse quantization on the input data, and the inverse orthogonal transform unit 106 performs inverse orthogonal transform processing by inverse discrete cosine transform on the inversely quantized data and is restored. Output image data.

  Next, the adder 108 adds the predicted image data output from the predicted image generation unit 107 described later to the image data output from the inverse orthogonal transform unit 106, and uses the added data as the memory A 110 and And output to the deblocking filter processing unit 109. The data after the addition is so-called local decoded image data.

  The deblocking filter processing unit 109 performs smoothing processing on the boundary portion of the encoded block included in the local decoded image data, and outputs the image data after the deblocking filter processing to the memory B111. Here, the memory A110 stores image data before the deblocking filter processing, and the memory B111 stores image data after the deblocking filter processing.

  The reduced image generation unit 112 thins out pixels by using a reduction method by configuring one pixel with an average value of a plurality of pixels with respect to an encoding target image included in input image data input to the image encoding device 100. A reduced image of the encoding target image is generated by a reduction method or the like. Next, the pre-motion search unit 113 uses the reduced image data output from the reduced image generation unit 112 to calculate an approximate motion vector for each block by comparing a plurality of frames. Here, the approximate motion vector is also called a wide-area motion vector (global vector) of the screen. Further, the pre-motion search unit 113 includes a difference absolute value sum calculation unit (SAD calculation unit) 116, and the difference absolute value between the input image and another image adjacent to the current image to be encoded. The sum (SAD: Sum of Absolute Difference) is calculated. Then, the SAD value information calculated by the calculation is supplied to the quantization parameter control unit 115.

  Next, the motion search unit 114 determines a prediction method of the predicted image using the input image data input to the image encoding device 100 and the image data stored in the memory A110 and the memory B111. Further, the motion search unit 114 refers to the motion vector output from the pre-motion search unit 113, performs a more detailed motion search, and calculates a motion vector related to motion compensation prediction at the time of inter prediction encoding.

  The operation of the motion search unit 114 will be specifically described below. When performing intra prediction, a predetermined pixel block (16 × 16 pixels, 8 × 8 pixels, 4 × 4) is used by using input image data and image data before deblocking filter processing stored in the memory A110. The encoding efficiency for each predetermined prediction mode is obtained in units of pixels. On the other hand, when performing inter prediction, encoding efficiency is calculated | required using input image data and the image data after the deblocking filter process stored in memory B111. When searching for a reference image at the time of inter prediction, the motion vector output from the pre-motion search unit 113 is used to search in detail around the motion vector so that the entire image area is not searched for. Realize processing. And the prediction method with which the calculated | required encoding efficiency becomes optimal is determined. In the case of inter prediction, a motion vector is further calculated and output. In addition, when the input picture is an I picture on the assumption that intra prediction coding is performed, only intra prediction is adopted, and when the input picture is another picture (P picture or B picture), intra prediction and inter prediction are used. Among them, it is possible to select a prediction method that provides optimum coding efficiency. In this way, the motion search unit 114 selects a prediction method with the best coding efficiency, and outputs coding parameters such as an intra prediction mode and a motion vector.

  Next, the predicted image generation unit 107 generates predicted image data using the prediction method and coding parameters output from the motion search unit 114 and the image data stored in the memory A110 or the memory B111. The predicted image data is supplied to the difference calculation unit 101 and the adder 108 as described above.

  Here, the processing of the quantization parameter control unit 115 will be described. The quantization parameter control unit 115 selectively executes a feedforward method and a feedback method. Further, the quantization parameter control unit 115 has a configuration for storing SAD values corresponding to at least several frames and calculating a difference value of SAD. Then, the quantization parameter control unit 115 encodes the SAD value for the encoding target image (current image) output from the pre-motion search unit 113 and the encoded image having the same picture type as the stored encoding target image. A difference value from the SAD value is calculated to obtain an image change amount.

  Then, it is determined whether or not the calculated image change amount is equal to or greater than a predetermined threshold. If the image change amount is equal to or greater than the threshold, the quantization parameter control unit 115 determines an initial quantization parameter for the encoding target image by a feedforward method. The feedforward method is a method (first method) for determining a quantization parameter by predicting a generated code amount from information of an original picture. In the feedforward method, a predetermined table indicating the predicted code amount for each quantization parameter, which is statistically calculated from the SAD value calculated by the pre-motion search unit 113, is used. That is, the quantization parameter indicating the predicted code amount that is the closest to the target code amount is set as the initial quantization parameter in the feedforward method.

  On the other hand, if the image change amount is less than the threshold, the quantization parameter control unit 115 determines an initial quantization parameter for the encoding target image by a feedback method. The feedback method is a method (second method) for determining a quantization parameter by predicting a generated code amount of an encoding target image from an encoded image. In the feedback scheme, the average quantization parameter of an encoded image of the same picture type as the encoding target image is used as the initial quantization parameter.

  As described above, the quantization parameter control unit 115 adaptively switches the initial quantization parameter determination method according to the image change amount based on the SAD value, and outputs the determined initial quantization parameter to the quantization unit 103. Thereafter, the quantization parameter control unit 115 observes a difference value between the preset target code amount and the code amount actually generated by the entropy encoding unit 104. Then, while observing the difference value, quantization control is performed by changing the quantization parameter so that the generated code amount for the encoding target image approaches the target code amount.

  Next, a control method in the quantization parameter control unit 115 will be described with reference to the flowchart of FIG. FIG. 2 is a flowchart showing an initial quantization parameter determination method.

  In step S201, the quantization parameter control unit 115 calculates the absolute difference between the SAD value of the encoding target image and the SAD value of an encoded image of the same picture type as the encoding target image. The absolute difference value is a value representing the image change amount of the scene. Next, in step S202, the quantization parameter control unit 115 determines whether or not the absolute difference value calculated in S201 is equal to or greater than a predetermined threshold value. If it is determined in S202 that the absolute difference value is less than the threshold, the flow proceeds to step S203 for executing the initial quantization parameter determination method by the feedback method. On the other hand, if it is determined in S202 that the difference absolute value is greater than or equal to the threshold value, the flow proceeds to step S204 for executing the initial quantization parameter determination method by the feedforward method.

  In S203, the quantization parameter control unit 115 determines the average quantization parameter of the encoded image of the same picture type as the initial quantization parameter. Further, in S204, the quantization parameter control unit 115 selects a predicted code amount indicating a value closest to the target code amount from the table obtained from the statistical result based on the SAD value of the encoding target image. Then, a quantization parameter corresponding to the prediction code amount is selected. In step S205, the quantization parameter control unit 115 determines the quantization parameter selected in step S204 as an initial quantization parameter.

  In the above-described embodiment, the configuration in which the image change amount is determined by calculating the SAD has been described. However, instead of the value of the SAD, for example, the calculation result of SATD (Sum of Absolute Transformed Difference) is used as an index. Also good.

  In the above embodiment, the present invention is applied regardless of the picture type. However, the present invention may be applied only when the encoding target image is an I picture.

  As described above, according to the image encoding device of the present embodiment, the initial quantization parameter determination method is adaptively switched according to the image change amount (scene continuity) determined by the change in the SAD value. Thus, suitable quantization control can be performed.

  In the first embodiment, the configuration for obtaining the image change amount from the difference information of the SAD in the reduced image has been described. In the second embodiment, the configuration for obtaining the image change amount from the difference information of the variance value will be described. FIG. 3 is a block diagram illustrating a schematic configuration of an image encoding device 300 according to the second embodiment of the present invention. The configuration of the image encoding device 300 will be described below with reference to the block diagram of FIG. In FIG. 3, blocks denoted by the same reference numerals as those in FIG. 1 overlap with those described in the first embodiment, and thus detailed description thereof is omitted. In the following, the configuration of blocks different from FIG. 1 will be mainly described in detail.

  The variance value calculation unit 301 acquires the reduced image data output from the reduced image generation unit 112 and calculates the variance value of the encoding target image (current image). Here, the variance value is an average value obtained by squaring the average value and the difference value of each data, and represents the degree of dispersion of pixel values. The higher the variance value, the greater the dispersion of pixel values, indicating that the luminance change of the pixels is severe. As human visual characteristics, it is known that image deterioration is less conspicuous as an image has a more intense change in luminance. In view of this, a process for obtaining the degree of difficulty of encoding for each image block based on the variance value is generally performed. In the image encoding device 300, information on the dispersion value calculated by the dispersion value calculation unit 301 is supplied to the quantization parameter control unit 115.

  Next, processing of the quantization parameter control unit 115 will be described. The quantization parameter control unit 115 selectively executes a feedforward method and a feedback method. Further, the quantization parameter control unit 115 has a configuration for storing a dispersion value corresponding to at least several frames and calculating a difference value of the dispersion values. Then, the quantization parameter control unit 115 outputs the variance value for the encoding target image (current image) output from the pre-motion search unit 113 and the encoded image of the same picture type as the stored encoding target image. A difference value from the variance value is calculated to obtain an image change amount.

  Then, it is determined whether or not the calculated image change amount is equal to or greater than a predetermined threshold. If the image change amount is equal to or greater than the threshold, the quantization parameter control unit 115 determines an initial quantization parameter for the encoding target image by a feedforward method. In the feedforward method, a predetermined table indicating the prediction code amount for each quantization parameter, which is statistically calculated from the variance value calculated by the pre-motion search unit 113, is used. That is, the quantization parameter indicating the predicted code amount that is the closest to the target code amount is set as the initial quantization parameter in the feedforward method.

  On the other hand, if the image change amount is less than the threshold, the quantization parameter control unit 115 determines an initial quantization parameter for the encoding target image by a feedback method. In the feedback method, an average quantization parameter of an encoded image of the same picture type as that of the encoding target image is used as an initial quantization parameter.

  As described above, the quantization parameter control unit 115 adaptively switches the initial quantization parameter determination method according to the image change amount based on the dispersion value, and outputs the determined initial quantization parameter to the quantization unit 103. Thereafter, the quantization parameter control unit 115 observes a difference value between the preset target code amount and the code amount actually generated by the entropy encoding unit 104. Then, while observing the difference value, quantization control is performed by changing the quantization parameter so that the generated code amount for the encoding target image approaches the target code amount.

  As described above, according to the image coding apparatus of the present embodiment, the initial quantization parameter determination method is adaptively switched according to the image change amount (scene continuity) determined by the change in the variance value. Therefore, suitable quantization control can be performed.

DESCRIPTION OF SYMBOLS 100 Image coding apparatus 101 Difference calculation part 102 Orthogonal transformation part 103 Quantization part 104 Entropy coding part 105 Inverse quantization part 106 Inverse orthogonal transformation part 107 Predictive image generation part 108 Adder 109 Deblocking filter process part 110 Memory A
111 memory B
112 reduced image generation unit 113 pre-motion search unit 114 motion search unit 115 quantization parameter control unit 116 SAD calculation unit 301 variance value calculation unit

Claims (3)

An image encoding device for encoding a moving image,
Quantization means for quantizing data of an encoding target image included in the moving image;
A first method for predicting a generated code amount from information of an original image of the encoding target image and determining an initial quantization parameter; and an initial value by predicting a generated code amount of the encoding target image from an encoded image A quantization control means for selectively executing a second method for determining a quantization parameter, and controlling quantization in the quantization means according to the determined initial quantization parameter;
A calculation means for calculating an image change amount between the encoding target image and a predetermined encoded image;
The quantization control unit selects the first method when the image change amount calculated by the calculation unit is equal to or greater than a predetermined threshold, and the image change amount calculated by the calculation unit is less than the predetermined threshold. In this case, an initial quantization parameter for the encoding target image is determined by selecting the second method. Motion search means for performing motion search on the encoding target image;
Reduced image generation means for generating a reduced image of the encoding target image supplied to the motion search means;
A difference absolute value sum calculating means for calculating a sum of absolute differences between a reduced image of the encoding target image generated by the reduced image generating means and a reduced image of another image;
The calculation means acquires information on the value of the difference absolute value sum calculated by the difference absolute value sum calculation means, and calculates the difference absolute value sum between the encoding target image and the predetermined encoded image. The image encoding apparatus according to claim 1, wherein the image change amount is calculated by calculating a difference value between the image change amount and the image change amount. Motion search means for performing motion search on the encoding target image;
Reduced image generation means for generating a reduced image of the encoding target image supplied to the motion search means;
A variance value calculating means for calculating a variance value from the reduced image of the encoding target image generated by the reduced image generating means;
The calculation means obtains information on the variance value calculated by the variance value calculation means, and calculates the difference value of the variance value between the encoding target image and the predetermined encoded image. The image coding apparatus according to claim 1, wherein the image change amount is calculated.

Images