KR20040062733A - Bit rate control system based on object - Google Patents

Abstract

PURPOSE: A method and an apparatus for controlling an object-based bit rate are provided to assign macro block quantization parameters to an object region and a background region on the basis of a frame and control the bit rate. CONSTITUTION: A picture quantization parameter setup unit(303) receives a target bit quantity and an object segmentation result from a target bit setup unit(302), and sets a picture quantization parameter. A macro block quantization parameter setup unit(304) suitably assigns a quantization parameter according as a macro block to be quantized belongs in an object region or a background region. A quantizer(305) quantizes an input image according to the quantization parameter of the corresponding macro block. An output buffer(306) outputs the quantized image signal as a bit stream.

Description

BIT RATE CONTROL SYSTEM BASED ON OBJECT}

The present invention relates to a method of controlling a bit rate by allocating macroblock quantization parameters to an object region and a background region based on a frame with a small amount of computation when a multimedia image is transmitted to a network.

When transmitting video data including a large amount of information through a bandwidth-limited channel, the bit rate is controlled according to network conditions. Bit rate control for image data transmission is generally performed by a control technique for determining encoding parameters. In the bit rate control technique, a forward control technique that allocates a bit rate in consideration of characteristics of an input image and a parameter of an encoder are determined in consideration of characteristics of a rear end of a source encoder such as a buffer state transmission rate. There is a backward control technique.

Video encoders for low speed transmission media such as video communications and video phones should have excellent performance in terms of compression efficiency and low complexity of the encoder. Therefore, most of the bit rate control schemes for the video encoder for a low-speed transmission medium mainly use a backward technique for controlling the bit rate by adjusting the quantization parameter (QP) considering the state of the buffer.

1 is a schematic diagram illustrating a general bit rate control method, comprising: a characteristic analyzer 101 for receiving an input image and analyzing a characteristic; an encoder 102; a buffer 103 for outputting the encoded signal as a bit string; A target bit allocation setter 104 for setting target bits based on the state of the input video signal and the bit string output buffer, and a bit rate adjuster 105 for adjusting the coding rate with the assigned target bits.

The characteristic analyzer 101 analyzes the characteristic of the input image data and provides the characteristic to the target bit allocation setter 104, and the target bit allocation setter 104 provides the characteristic of the input image and the state information of the output buffer 103. It takes input and sets target bit. The bit rate adjuster 105 adjusts the encoding bit rate of the bit rate encoder 102 according to the target bit set in the target bit allocation setter 104, and the encoder 102 adjusts the image data that has passed through the characteristic analyzer 101. The data is encoded in accordance with the bit rate and stored in the output buffer 103. The output buffer regulates the output bit stream at a constant speed to transmit the variable bit stream through a channel having a limited bandwidth, and also prevents buffer overflow or buffer deficiency that may occur during bit stream recovery. do.

In general, the adjustment of the bit rate can be adjusted by using a coding parameter such as a quantization parameter (QP). For example, when the quantization parameter value is set large, the bit rate is lowered but the image quality is lowered. The higher the bit rate, the better the picture quality. As described above, since the image quality and the bit rate are in mutual relation with each other, the image quality deteriorates when the bit rate drops.

The most well-known method of bit rate adjustment using quantization parameters is the bit rate adjustment method used in the TMN used as a test model in the standard group that established H.263.

The TMN is followed by a number after the TMN, depending on the version of the test model, of which TMN10 uses a variety of information generated from the encoder as a bit rate control method used in H.263 +. 2 is a schematic diagram of a TMN10 bit rate adjustment method.

The encoding initialization step S201 initializes the buffer size and the buffer level and initializes the parameters used in the bit rate adjustment model. In the frame layer bit rate adjustment step (S202), the frame is skipped until the frame is smaller than the reference bit rate in order to check the buffer state to prevent the buffer overflow and delay. For reference, the TMN10 bit rate control method uses a small buffer because it is intended for real-time transmission.

In the MB (macroblock) layer bit rate adjustment steps (S203 and S205), the number of frame bits to be encoded is predicted based on the bit rate adjustment model, and the encoding is performed while performing the bit rate adjustment by adjusting the QP in MB units. Finally, the parameters used as the bit rate adjustment model are updated (S204).

However, the TMN10 bit rate adjustment method uses a parameter that requires a large amount of computation, such as the variance between this frame and the current frame, in using the bit rate adjustment model. In addition, since the quantization parameter is adjusted in MB units, a user may feel awkward due to the heterogeneity of the image quality of a part having good image quality while the other part has a poor image quality even though the same object area is in one frame.

However, the bit rate control method as described above presents one mathematical model (linear model, nonlinear model, replicalin model, exponential model, Gaussian model, etc.) for all images and assigns bits accordingly. However, there is a disadvantage in that an optimal coding parameter cannot be determined for an important part of the input image. In other words, if the bit rate is lowered due to the deterioration of the network environment even in an important area, the image quality may be deteriorated, and even in an unimportant area, the image quality may be higher than that of the important area.

Such correlation between bit rate and image quality is considered in video communication and video phone communication. That is, the user reacts more sensitively to the deterioration of the image quality of the important area or the region of interest than the deterioration of the image quality of the non-region. In other words, the image quality of the important area or the region of interest is greatly influenced by the user's image quality rather than the image quality of the entire screen. However, in the general bit rate control method, the quantization parameter is adjusted according to the overflow of the output buffer or the lack of the output buffer. Therefore, the image quality is large and the quality deterioration occurs frequently according to the network conditions, so that the user cannot provide sufficient image quality.

An object of the present invention is to provide a bit rate control method and apparatus capable of eliminating the heterogeneity of high quality and high quality of the bit rate control method provided by the above-described TMN10.

In particular, the present invention solves the disadvantages of the conventional bit rate control scheme described above, and adaptively controls the bit rate in a network environment that changes with a small amount of computation, while appropriately discriminating quantization parameters of important and non-critical areas. It is an object of the present invention to provide a bit rate control method and apparatus for improving subjective picture quality over a general method in the case of the same bit rate by first assigning the picture quality in a non-critical area.

1 is a schematic diagram of a general method of controlling a bit rate

Figure 2 is a schematic diagram of the TMN10 bit rate adjustment method

3 is a schematic diagram of a bit rate control apparatus according to the present invention;

4 is a flowchart of a bit rate control method according to the present invention;

5 is a flowchart of a method for generating a picture quantization parameter PQUANT in the present invention.

6 is a flowchart of a method for assigning quantization parameters to respective regions in the present invention.

7 illustrates an example of object segmentation;

In order to achieve the above object, the object-based bit rate control method of the present invention comprises: setting a target bit of a video signal to be transmitted through a network, setting a quantization parameter (PQUANT) in units of pictures using the set target bit, Quantizing the macro block by applying the quantization parameter PQUANT of the set picture unit; Characterized in that it comprises a.

In addition, the object-based bit rate control apparatus of the present invention for achieving the above object, the means for setting the target number of bits for bit rate calculation and control according to the network environment, means for separating the input image into the object area and the background area, Means for updating a picture quantization parameter PQUANT in consideration of the set target number of bits, and assigning a quantization parameter value ROI_QP of an object region and a quantization parameter value BG_QP of a background region based on the picture quantization parameter PQUANT. Means for quantizing an input image by applying a corresponding quantization parameter value for each macroblock corresponding to each of the allocated regions; Characterized in that comprises a.

The bit rate control method of the present invention made as described above will be described in detail with reference to the accompanying drawings.

The bit rate adjustment algorithm, which is common in network communication such as video communication and video phone, is not suitable for use in a wireless terminal because of its large amount of computation. Therefore, when the application to a wireless terminal such as IMT2000, considering the amount of calculation, a small amount of algorithm is required. Accordingly, in the present invention, a picture quantization parameter is generated using a bit rate adjustment algorithm having a small amount of computation. The input image is segmented to generate object region (ROI) information, and the macroblock quantization parameter (QP) is based on the picture quantization parameter, and the object quantization parameter (ROI_QP) is used by using 'AnnexT Modified Quantization Mode'. ) And background quantization parameters (BG_QP).

As described above, in the conventional bit rate control method in the TMN10, the bit rate is controlled by first determining the frame level QP and the macro block level QP in consideration of the network environment and the current buffer situation. In the present invention, only the QP of the frame level is determined for a simple calculation amount, and then the QP of the macroblock level is determined based on the ROI region based on the determined frame QP. That is, the QPs allocated to the macroblocks corresponding to the ROI region in one frame are all constant, and the QPs allocated to the macroblocks corresponding to the region other than the ROI are also constant.

3 shows a schematic diagram of a technique for controlling a bit rate by adjusting a quantization parameter value based on an object of the present invention. The target bit setter 302 sets the target bit amount from the result 301 of performing object segmentation and the result of checking the state of the output buffer 306. The picture quantization parameter setter 303 sets the picture quantization parameter PQUANT by inputting the target bit amount and the object segmentation result set by the target bit setter 302. The macroblock quantization parameter setter 304 appropriately allocates the quantization parameter MB_QP according to whether the macroblock to be quantized belongs to the object region or the boundary of the background region. As described above, the quantization parameter assigned to the macroblock corresponding to the object region ROI based on the quantization parameter of the frame level, that is, the picture quantization parameter PQUANT, is uniformly allocated, and the non-ROI region (background region: The quantization parameter allocated to the macroblock corresponding to BG) is also constantly assigned.

The quantizer 305 quantizes the input image according to the quantization parameter MB_QP of the corresponding macroblock set according to the background region, the object region, etc., and the output buffer 306 outputs the quantized image signal as a bit stream. Done.

In the bit rate control method using the quantization parameter of FIG. 3, when performing multimedia communication through a network, a target bit is set based on a result from an output buffer, a quantization parameter is set based on a frame, and then based on an object segmentation result. As a technique of specifying different quantization parameters in macroblock units according to an important region and an unregional region of an image, it is based on the following technique.

According to a moving picture standard such as H.263 or MPEG1 / 2, each frame of a given picture is encoded in macroblock units, which are 16 × 16 sub-regions. The brightness component and the color component of each macro block go through a discrete cosine transform (DCT) and quantize coefficient values to represent N bits. The quantized value is again encoded by Variable Length Coding (VLC). At this time, determining the method of performing the quantization is a quantization parameter, the number of coefficients smaller than the quantization parameter is '0'. Therefore, if the value of the quantization parameter is large, the coefficient converted to '0' is increased, resulting in a loss of information due to quantization instead of a small number of bits as a result of encoding by the VLC. Information loss due to quantization is reduced and image quality is improved, but the burden of data transmitted over a limited bandwidth network is increased.

Accordingly, in the present invention, the quantization parameter value of the frame level is determined in the image to be transmitted, and the image quality is increased by setting the quantization parameter value relatively small for the region of interest or importance, and the region for the region of interest or non-importance is determined. It is proposed a technique to reduce the burden of data transmission by setting a relatively large quantization parameter value.

The segmentation result 301 for the input image provides information on important and non-important regions of the input image. Object segmentation is a technique of separating an object from an image input image, and is a technique of separating each object when there is a user and a background or several objects. When such a technique is provided, the separated object and the background area have different importance in separating the objects. Subjective picture quality, which is usually felt by a person, is object dependent rather than background, and especially in a video communication system, the user (object) image contributes more to recognition than the background image. Allocating relatively fewer bits to the background can improve subjective picture quality.

Therefore, based on the information imparted with importance in the object region and the background region, the quantization parameter setter 303 at the picture level and the quantization parameter setter 304 at the macroblock level have different quantization parameter values for each region (macro block). By properly setting the quantizer 305, the quantizer 305 performs quantization based on this, thereby improving the overall subjective image quality by increasing the image quality of the region that is important to the user and lowering the image quality of the region that is not of interest. Will be.

Fig. 4 determines the quantization parameter value at the frame level and classifies it into an important region (in this case, a macro block of the object region) and an insignificant region (in this case, a macro block of the background region) based on the object segmentation result. A method of allocating a quantization parameter based on the quantization parameter of the frame level is shown.

In step S401, object segmentation information (ROI information) is input. The next step S402 is to determine the quantization parameter value of the frame level.

The next step (S403) is a step of determining whether the ROI region based on the object segmentation result. If the result of the determination is a ROI region, the process proceeds to step S404a and all constant quantization parameters are assigned to the macroblocks corresponding to the ROI region, and if not, the process proceeds to step S404b to the macroblock corresponding to the BG region. All assign a constant quantization parameter. The next step S405 is to quantize the macro block based on the quantization parameter assigned to the macro block for each region of the frame quantization parameter reference.

As described above, in the present invention, the quantization parameter value of the frame reference is assigned and the bit rate control is performed by adjusting the importance-based quantization parameter value of the object / background region based on the reference.

In a video communication environment, an important area generally may be a human area or a face area or an eye or mouth area, and may be composed of one or two or more thereof. In addition, in designating the importance of the object, the importance of the human area, the face area, and the eye or mouth area may be set higher than that of the background area.

Object separation is a process of separating a human and a background area or a face and a background area. In fact, when the face region appears in high quality, the user may not feel the quality deterioration even if the other peripheral region is degraded by the quantization parameter. This is because the user's eyes are mainly focused on the face part of the image, so the sensitivity of the image quality is also focused on the face part.

Therefore, the subjective picture quality can be increased by classifying the quantization parameters into important and non-important areas.

In other words, the basic concept of the bit rate adjustment method shown in FIG. 4 is to control the bit rate by allocating frame-based quantization parameter values. The smaller and less important parts are larger quantization parameters, so that important objects can be assigned a larger number of bits even if the bit rate is lower, while non-critical objects can be lowered as much as possible. The other part is coded separately. This makes the user feel less deterioration in image quality. That is, the encoder separates the important areas from the object segmentation results into macroblock units and performs encoding according to the object importance-based quantization parameter value, thereby reducing the image quality while maintaining the target number of bits, and limited network. The bandwidth burden can be solved.

The method of determining the quantization parameter value (QP) of the frame level in the bit rate control method of the present invention and a detailed method of correcting the QP of the ROI region and the remaining region based on the determined QP will be described.

First, a method of generating picture quantization parameters will be described.

According to a video standard such as H.263 or MPEG1 / 2, the quantization parameter of a given picture has a PQUANT and GOB or slice structure that determines the quantization parameter of the entire picture, which is the highest layer according to the coding structure. There is a corresponding GQUANT, SQUANT. And finally there is DQUANT which determines the quantization parameter of the macro block.

Quantization is performed using the quantization parameter generated in each layer of the quantization parameter. When quantization is performed in each substructure with a different quantization parameter, the quantization parameter can be changed using the corresponding QOUNT.

5 is a schematic diagram of generating a quantization parameter in a picture layer proposed in the present invention. First, the buffer state is checked by using the method used in the TMN10 bit rate adjustment method to prevent overflow and to prevent delay.

That is, the first step (S501) calculates the current number of bits (Curr_bits) as one frame encoded bits, the target number of bits (target_bits) as bit rate (bit_rate) / target frame rate (target_frame_rate), Compute the current buffer status (CommBuffer) as CommBuffer = CommBuffer + Curr_bits-target_bits. In operation S502, when the current buffer state CommBuffer is larger than the target bits by comparing with the target bits target_bits, the process proceeds to step S503 to empty the buffer by the number of target bits and skip the frame.

After performing the frame skip process, the PQUNT is updated (S504 and S505). The basic concept is to compare the current encoded bit amount (Curr_bits) with the target bit amount (target_bits) and increase or decrease the quantization parameter if the encoded bit amount is large. Depending on the amount of bits encoded, the threshold representing the quantization parameter increment is an experimental value and may vary depending on the system to be applied.

The process of generating the picture quantization parameter PQUANT after performing a frame skip process will be described in more detail. The current encoded bit amount Curr_bits is compared with the target bit amount target_bits.

If Curr_bits> target_bits * 3 indicates that the quantization parameter scale value (QP_scale) = 9 and Curr_bits> target_bits * 2 indicates that the quantization parameter scale value (QP_scale) = 6 and Curr_bits> target_bits * 1.5, the quantization parameter scale value (QP_scale) = 4 If Curr_bits> target_bits, then quantization parameter scale value (QP_scale) = 2, Curr_bits> target_bits * 0.9, if quantization parameter scale value (QP_scale) = 0, Curr_bits> target_bits * 0.7, quantization parameter scale value (QP_scale) = -2, Curr_bits> If target_bits * 0.5, the quantization parameter scale value (QP_scale) = -3, and Curr_bits <target_bits * 0.5, the quantization parameter scale value (QP_scale) = -3, respectively.

That is, the meaning of step S504 is that the current encoded bit amount Curr_bits and the target bit amount target_bits are compared, and if the encoded bit amount is greater than three times the target bit amount, the value is 9 by the quantization parameter of the previous frame. (PQUANT) is increased, and if the encoded bit amount is greater than 2 times and less than 3 times the target bit amount, the value PQUANT is increased by 6 than the quantization parameter of the previous frame, and the encoded bit amount is 1.5 times the target bit amount. If it is greater than 2 times and less than 2 times, the value (PQUANT) is increased by 4 than the quantization parameter of the previous frame.If the encoded bit amount is larger than the target bit amount and less than 1.5 times, the value (PQUANT) is increased by 2 than the quantization parameter of the previous frame. PQUANT), and if the encoded bit amount is smaller than the target bit amount and larger than 0.9 times, the quantization parameter of the previous frame is maintained, and the encoded bit amount is the target bit. Is less than 0.9 times and greater than 0.7 times, then decreases its value (PQUANT) by 2 than the quantization parameter of the previous frame, and if the encoded bit amount is less than 0.7 times and greater than 0.5 times the target bit amount, If the value PQUANT is decreased, and the encoded bit amount is less than 0.5 times the target bit amount, it means that the value PQUANT is decreased by 5 than the quantization parameter of this frame.

To this end, in the next step S505, the current picture quantization parameter value Curr_PQUANT is updated in consideration of the quantization parameter scale value QP_scale and the previous picture quantization parameter value prev_PQUANT. In a next step S506, ROI_QP and BG_QP are allocated using the updated current picture quantization parameter value Curr_PQUANT.

As described above, the quantization parameter scale value QP_scale is a value representing a quantization parameter increase width according to the encoded bit amount, and the threshold is set to an experimental value.

In addition, depending on the application, the compromise between spatial and temporal image quality may be changed. In this case, the spatial image quality refers to the image quality that is felt when viewing a frame of the video, and the temporal image quality refers to the extent to which the motion of the image is naturally felt when the video is played. For example, if the number of frames displayed per second is small, the movement is not natural and thus the temporal image quality deteriorates. Therefore, if spatial quality is important, even if frame skipping occurs, one frame is encoded with high quality. If temporal image quality is important, even if the quality of one frame is somewhat reduced, the frame skipping is encoded as much as possible.

Next, a method of allocating ROI_QP and BG_QP using Curr_PQUANT, that is, a method of allocating quantization parameters to respective regions will be described.

A method of classifying quantization parameters into important and non-important areas and assigning them to each other is schematically illustrated in FIG. 6. First, an important region quantization parameter (ROI_QP) value and an insignificant region quantization parameter (BG_QP) value are allocated using the picture quantization parameter value. The picture quantization parameter value is divided into several regions according to the size, and ROI_QP and BG_QP values are determined according to the range of the quantization parameter value.

This process is described in more detail.

First, ROI separation information is input from the object segmentation result (S601), and then the quantization parameter value QP_new is calculated using the TMN10 bit rate adjustment algorithm (S602). In a next step S603, the quantization parameter ROI_QP of an important region (object region) and the quantization parameter BG_QP of an insignificant region (background region) based on the quantization parameter value QP_new calculated by the TMN10 bit rate control algorithm. ) Will be assigned.

Here, it is divided into several areas according to the size of the quantization parameter value (QP_new) calculated by the TMN10 bit rate adjustment algorithm. That is, when QP_new> 25, 20 <QP_new <25, 10 <QP_new <20, QP_new <10.

The ROI_QP and BG_QP values are determined in the next step S604 according to the range of the quantization parameter value QP_new calculated as TMN10. In the method of allocating ROI_QP and BG_QP, weights are set according to importance, and usually ROI_QP is set at about half of QP_new and BG_QP is set at 1,5 times. However, if QP_new exceeds 25, lowering ROI_QP as usual would result in a lot of frame skipping due to a sudden increase in data, resulting in poor temporal image quality. Therefore, in this case ROI_QP prevents frame skipping by assigning a relatively high quantization parameter. In addition, when QP_new is less than 10, even if the quantization parameter is further lowered, the degree of image recognition that humans can recognize is very small, so in this case, the overall image quality can be maintained by increasing the quality of non-essential areas.

That is, if QP_new> 25, ROI_QP = QP_new * 0.8 (weighted), BG_QP = 31, and if 20 <QP_new <25, ROI_QP = QP_new * 0.5 (weighted), BG_QP = 31, and 10 <QP_new < In case of 20, ROI_QP = QP_new * 0.5 (weighted) and BG_QP = QP_new * 1.5 (weighted), and when QP_new <10, ROI_QP = QP_new and BG_QP = QP_new.

The next step S605 is adjusting the quantization parameter value. That is, the quantization parameter dquant of the macro block is adjusted by dquant = prev_QP-(ROI_QP or BG_QP). Where prev_QP is the previous quantization parameter value, ROI_QP is the set object quantization parameter value, and BG_QP is the set background quantization parameter value.

Next steps S606a and S606b are steps of adjusting the dquant value according to the macroblock type, Pre_MBType represents the previous macroblock quantization type, and Cur_MBType represents the current macroblock quantization type. The next step (S607a) is to use the previous quantization parameter value (prev_QP) as ROI_QP or BG_QP when dquant is in the range ± 3, and step S607b refers to the current ROI_QP or BG_QP when dquant> 3 or dquant <3 It means to use.

The next step S608 is to perform final quantization using the quantization parameter set as described above.

In the present invention, H.263 + AnnexT (Modified Quantization Parameter) is used to greatly change the quantization parameters of ROI_QP and BG_QP. In the existing H.263, the quantization parameter can be adjusted from +2 to 2 of the quantization parameter value of the previous macroblock in macroblock units. That is, the degree of change of the quantization parameter for changing the bit rate is possible only within ± 2 of the value of the quantization parameter in the previous macro block. As a result, it is difficult to adapt to sudden network environments, and it is difficult to make a significant change between major and non-primary domains. However, in Annex T, the quantization parameter is adjusted using two bits in 'small-step QUANT alteration mode' using the difference of the specified QUANT as shown in Table 1, or 6 bits in 'arbitrary QUANT selection mode'. Regardless of whether the quantization parameter value is adjusted from 1 to 31, it has been extended.

Using this function, if the network's bit rate is lowered, the DCT coefficients of macro blocks determined to be insignificant areas are quantized to a large value to reduce the amount of data, and the DCT coefficients of macro blocks determined to be important areas are quantized to small values. Thus, even when the bit rate is low, relatively good image quality can be maintained for the region of interest.

When the user separates an image based on an object and tries to quantize differently by determining the importance, an easy and fast way to adjust the bit rate is to use the 'arbitrary QUNAT selection mode'. 7 shows an example of object segmentation and shows that it can be divided into an object region and a background region. As shown in FIG. 7, when the object segmentation is performed in the image communication and the background region and the user region are separated, the two portions may be separated and encoded using different quantization parameters.

In case of using 'Arbitrary QUANT selection mode', users can select any quantization parameter from 1 to 31 for each macroblock regardless of the quantization parameter value of previous macroblock. The bit rate can be adjusted while quantizing the object region and the background region differently. If the image is divided into an object and a background region as shown in FIG. 7, the user region (object region) is quantized using a small quantization parameter to leave much DCT coefficient information when the bit rate is controlled using this information. The bit rate is controlled by quantizing using a relatively large quantization parameter to discard DCT coefficient information. In this way, areas of interest or interest can be represented with relatively good image quality, while areas of interest or relative interest are notably low.

However, when both areas use the 'arbitrary QUANT selection mode' of Annex T, the number of header bits added increases, which may be a burden on the bit rate control. In other words, when using a quantization parameter in a general manner, a 2-bit header is required per macroblock. When using an 'arbitrary QUANT selection mode' of Annex T, a 6-bit header is required. Since this additional bit is added for each macro block, the overhead may be large.

However, when the image is separated and input based on the object, the difference between the quantization parameter is large only in the boundary region (boundary macroblock) of the background region and the object region, and the difference in the quantization parameter is small within each region. It will be very different. When the object and the background are separated from the image using this property, the arbitrary quantization parameter desired by the user is determined using the 'arbitrary QUANT selection mode' only for the macroblock of the object's boundary region, and in the same region, the previous quantization parameter is determined. If the difference in value is within the range of ± 3, the same quantization parameter is used so that 6-bit headers are used only for the border region and 2 bits are used for the rest, so that the header bits are more than 6-bit headers for the entire frame. You can adjust the bit rate on a per-object basis with much reduction.

As described above, if the current macroblock is the boundary macroblock of an object when quantizing based on the importance information imposed on the macroblock basis in the bit rate control, the quantization parameter is greatly adjusted using the 'arbitrary QUANT selection mode' and the current macro If the block is a macro block within an object, the quantization parameter can be adjusted to a small width using the 'small-step QUANT alteration mode', so that the number of bits can be adjusted in units of objects. By imposing a large number of bits, even in low bit rate transmissions, the image quality can be prevented while adjusting the bit rate.

In FIG. 7, the boundary area using the arbitrary QUANT selection mode is a portion denoted by a dot, and the macroblock or the important area in the macroblock coding sequence that passes from the non-essential area to the important area when viewed in the macroblock coding order is not important. It becomes a macro block that goes into an area. The coding order of a macroblock of one frame is made from the upper left to the lower right in line units.

The present invention can also control spatial and temporal image quality, which depends on how many bits are allocated to the critical area. In other words, if ROI_QP is made smaller, the spatial quality of the region of interest is increased, but the probability of frame skipping is relatively increased. Therefore, we propose a quantization parameter value that best satisfies the spatial and temporal image quality. Table 2 shows each quantization parameter value according to the range of QP_new for this purpose.

QP_new> 25 20 <QP_New <25 15 <QP_new <20 10 <QP_new <15 QP_new <10 ROI_QP 25 15 13 10 8 BG_QP 31 31 25 25 20

As shown in Table 2, in the present invention, the quantization parameter value QP_new calculated according to the TMN10 bit rate adjustment algorithm is divided into five regions (ranges), and the object region quantization parameter ROI_QP and the background region quantization parameter BG_QP are respectively determined. Assigned. Referring to Table 2, ROI_QP = 25 and GB_QP = 31 for QP_new> 25, and ROI_QP = 15 and BG_QP = 31 for 20 <QP_New <25, and ROI + _QP = for 15 <QP_New <20. 13, BG_QP = 25, ROI_QP = 10, BG_QP = 25 when 10 <QP_New <15, ROI_QP = 8, BG_QP = 20 when QP_new <10.

Until now, the present invention has described a method for adaptively allocating quantization parameters to important and non-critical areas by adjusting quantization parameters (QP) on a frame basis in order to effectively control bit rate in a real-time communication environment.

The present invention proposes a ROI-based bit rate control method capable of controlling the bit rate with a small amount of computation in a device having limited resources such as a mobile terminal.

The present invention is an algorithm for adaptively assigning quantization parameters to each region based on information about an important region and an insignificant region of an image segmented in units of objects. As a result, the image quality that a person feels at the same bit rate can be improved.

In addition, the present invention provides a visually cleaner feeling because the image quality difference does not occur in the same ROI region in one frame, and despite the control of the bit rate of the frame unit, a method of controlling the bit rate by the macro block unit by the conventional complex calculation Compared to this, the network utilization and frame rate are not reduced.

According to the present invention, when a video is segmented based on an object in a video communication or video phone transmission, the user can greatly set the importance of a specific object, thereby improving the image quality of the video because the user intuitively feels in the video communication. have.

In low bit rate communication, the quality of the transmitted video is not constant and the quality deterioration occurs frequently depending on the bandwidth of the network. This is because when the bit rate is controlled, frame skipping occurs to control the overflow of the output buffer, or the instantaneous bit rate is higher than the object in the video. This is because the quantization parameter is determined accordingly. However, in video communication or video phone communication, the user is sensitive to the image quality of the region of interest. Therefore, when controlling the bit rate, sending more information of the region of interest and less information of the region of interest is useful for controlling the bit rate. But it can provide a satisfactory picture quality to the user. In order to solve the problem caused by the calculation amount when the basic bit rate adjustment algorithm is applied to an application such as a wireless terminal, the present invention brings a large amount of calculation reduction and maintains the bit rate adjustment performance.

The present invention can be particularly useful in wireless mobile communication, such as IMT 2000. Since wireless mobile communication uses a lower network environment than wired mobile communication, it is very important to obtain high image quality at a low bit rate. In addition, it is possible to improve the performance of the terminal by reducing the power consumption by reducing the amount of computation.

Claims (13)

Setting a target bit of a video signal to be transmitted through a network, setting a quantization parameter PQUANT in picture units using the set target bit, and applying a corresponding quantization parameter PQUANT in the picture unit Quantizing; Object-based bit rate control method comprising a. 2. The method of claim 1, wherein frame skipping is performed to prevent buffer overflow and delay by using a currently encoded bit amount, a target bit amount, and a channel capacity. The object of claim 1, wherein the picture quantization parameter PQUANT is updated according to a difference between an encoded bit amount and a target bit amount obtained by comparing a current encoded bit amount with a target bit amount and comparing the bit amounts. Based bit rate control method. The method of claim 1, further comprising comparing the currently encoded bit amount with the target bit amount, If the encoded bit amount is larger than three times the target bit amount, the value PQUANT is increased by 9 than the quantization parameter of the previous frame. If the encoded bit amount is greater than 2 times and less than 3 times the target bit amount, increase the value PQUANT by 6 than the quantization parameter of the previous frame, If the encoded bit amount is greater than 1.5 times and less than 2 times the target bit amount, increase the value PQUANT by 4 than the quantization parameter of the previous frame, If the encoded bit amount is larger than the target bit amount and less than 1.5 times, increase the value PQUANT by 2 than the quantization parameter of the previous frame, If the encoded bit amount is smaller than the target bit amount and larger than 0.9 times, the quantization parameter of the previous frame is maintained. If the encoded bit amount is smaller than 0.9 times the target bit amount and larger than 0.7 times, the value PQUANT is decreased by 2 than the quantization parameter of the previous frame. If the encoded bit amount is less than 0.7 times the target bit amount and larger than 0.5 times, the value PQUANT is decreased by 3 than the quantization parameter of the previous frame. And if the encoded bit amount is less than 0.5 times the target bit amount, the value PQUANT is reduced by 5 than the quantization parameter of this frame. Setting a picture quantization parameter (PQUANT) to adjust the bit rate of an image signal to be transmitted through a network, and using the ROI separation information based on the set picture quantization parameter (PQUANT), an important region (ROI) and an insignificant region Specifying (BG) in units of macroblocks and setting macroblock quantization parameters (ROI_QP, BG_QP) differently according to importance. 6. The method according to claim 5, wherein the bits are allocated to areas that are more important than areas that are not important according to object importance by setting the quantization parameter differently according to the importance of the macroblock. [6] The method of claim 5, wherein the method for setting the quantization parameter for bit allocation of the macroblocks of the critical and non-critical regions according to the ROI information uses AnnexT (Modified Quantization) of H.263 +. Object based bit rate control method. The method of claim 5, wherein when setting the quantization parameter value of the macroblock, if the current macroblock is a boundary area between the background and the object, the quantization parameter value is changed using the 'arbitrary QUANT selection mode' in the AnnexT mode of H.263. And performing quantization using the same quantization parameter value if the current macroblock is the same region. The method according to claim 5, wherein after the picture quantization parameter PQUANT is updated, a quantization parameter value QP_new is calculated using a TMN10 bit rate control method. If set to 20 <QP_new <25, set ROI_QP = QP_new x 0.5, and BG_QP = 31, and if 10 <QP_new <20, set ROI_QP = QP_new x 0.5, and BG_QP = QP_new x 1.5, and QP_new <10, wherein ROI_QP = QP_new and BG_QP = QP_new. The method according to claim 5, wherein after updating the picture quantization parameter PQUANT, a quantization parameter value QP_new is calculated using a TMN10 bit rate control method, and when QP_new> 25, ROI_QP = 25 and BG_QP = 31. When ROI_QP = 15 and BG_QP = 31 when 20 <QP_new <25, set ROI_QP = 13 and BG_QP = 25 when 15 <QP_new <20 and ROI_QP when 10 <QP_new <15 = 10, BG_QP = 25, and when QP_new <10, the ROI_QP = 8 and BG_QP = 20. Means for setting a target bit number for calculating and controlling a bit rate to be transmitted according to a network environment, means for separating an input image into an object region and a background region, and means for updating a picture quantization parameter PQUANT in consideration of the set target bit number Means for allocating a quantization parameter value (ROI_QP) of an object region and a quantization parameter value (BG_QP) of a background region based on the picture quantization parameter (PQUANT), a corresponding quantization parameter for each macroblock corresponding to each of the allocated regions. Means for quantizing an input image by applying a value; Object-based bit rate control device comprising a. 12. The method of claim 11, wherein whether the current macroblock is an area boundary is determined based on the separation information of the object area and the background area, and the allocated quantization parameter is assigned to macroblocks that are area boundaries and macroblocks that are not. And means for selectively differently applying a quantization mode according to a value. 12. The apparatus of claim 11, wherein the quantization mode applied to the macroblock that is the region boundary is an arbitrary QUANT selection mode in AnnexT of H.263 +.