Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
  • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
  • H04N19/51—Motion estimation or motion compensation

Definitions

• This invention relates to apparatus for encoding a video signal representing the intensity of a plurality of picture elements comprising a predictor for predicting the intensity values of picture elements in a present frame of the signal using estimates of displacement of objects in the picture between the present frame and a prior frame.
• the picture information is encoded by identifying
• displacement is estimated recursively, using the previous estimate to form successive updates or correction terms.
• applicants estimation apparatus is not readily compatible with the prior art encoder because that prior art encoder requires a single displacement estimate per field, frame or block of data.
• the problem is solved in accordance with the invention with apparatus for encoding a video signal representing the intensity values of a plurality of picture elements in which the apparatus further comprises delay circuitry for storing an initial estimate of the displacement, and updating circuitry responsive to the storage means for repeatedly updating the stored estimate.
• the prediction error value is encoded for transmission.
• intensity data need not be encoded for pels in the background and the compensable areas, since this information can be accurately reconstructed in the decoder.
• FIG. 1 is a block diagram of a motion compensated encoder constructed in accordance with the present invention
• FIG. 2 is a block diagram of segmenter 112 of FIG. 1;
• FIG. 3 is a block diagram of formatter 140 of FIG. 1;
• FIG. 4 is a block diagram of segment detector and encoder 320 of FIG. 3;
• FIG. 5 is a block diagram of a second motion compensated encoder embodiment in which displacement estimates are encoded;
• FIG. 6 is a block diagram of apparatus for. decoding the output of the encoder of FIG. 1;
• FIG. 7 is a block diagram of deformatter 602 of FIG. 6; and
• FIG. 8 is a block diagram of segment detector and decoder 704 of FIG. 7.
• FIG. 1 A block diagram of a first embodiment of amotion compensated video encoder constructed in accordance with the present invention is shown in FIG. 1.
• a displacement estimator and an associated predictor together designated 100, which forms on line 151 a prediction p of the intensity of a moving area in the present frame in accordance with a displacement estimate which describes the location of the same area at a previous time.
• the displacement estimates generated by the apparatus of FIG. 1 are recursively updated, and, for this purpose, a one element delay circuit 104 is provided to supply the previous estimate to one input of an adder 110 which receives its second input from displacement error calculator 109.
• the latter provides a displacement error or update of the i-1 th estimated displacement during the time interval [t-t ,t], which is proportional to ( ) , , where is the displaced frame difference, i.e., the intensity difference between vector location X a (the picture element which is being processed) in the present frame, and a location displaced from X a by vector in the previous frame, and where VI is the intensity gradient measured at the displaced location in the previous frame. is the truncated or rounded value of which simplifies calculation of the gradient by interpolation of stored intensity values.
• the output of adder 110 is the new displacement estimate also a vector quantity.
• Displacement estimates are updated only in the picture area in which there is movement, and, for this purpose a switch 111 is interposed between the output of calculator 109 and adder 110. Switch 111 is closed under the control of a segmenter 112, similar to segmenter 312 of the copending application, but which performs additional functions as described below. Segmenter 312 receives the input video signal on line 101.
• the displacement estimate output from delay circuit 104 is input to first and second interpolators 131 and 107, and to a quantizer 105 which corresponds to quantizer 305 of the copending application.
• This quantizer can either round or truncate the displacement estimate to an integral value which is then applied to the address input of a frame memory 102.
• Memory 102 then returns, on lines 106, a series of stored intensity values for picture elements in the neighborhood of the location specified by Memory 102 may be a tapped delay line and associated control circuitry arranged to derive outputs from appropriate taps, as will be well known to those skilled in the art
• the intensity values from memory 102 and the displacement estimate from delay circuit 104 are used by interpolator 107 to form two quantities, namely the intensity values X at the picture location specified by the displacement estimate, and the intensity gradient at the displaced location (after rounding or truncation) at time t- t , when T is the time period between the present frame and the prior frame stored in memory 102. Both calculations may be made by straightforward linear interpolation.
• the gradient value is coupled directly to calculator 109, while the intensity value is applied to one input of a subtractor circuit 103, the other input of which is the intensity value input to frame memory 102, but delayed by one line period in a delay circuit 132.
• the line delay is provided because a new intensity value for the presently processed pel has not yet been entered in memory 102; the value of the immediately preceeding pel could have been used, by replacing the line delay with an element delay.
• this alternative would place processing time limitations on the apparatus which can be avoided by use of the longer line interval.
• the output of subtractor circuit 103 is the displaced frame difference DFD at the location specified by and this value is also applied to calculator 109, for use in formation of the update termgiven in equation (1) above.
• the preceeding description of the displacement calculation can be summarized by noting that a displace ment estimate is formed for each picture element in the moving area of the picture, and each estimate is updated using the previous estimate.
• the feedback used to enable the recursion is provided by the displaced frame difference and an intensity gradient value input to calculator 109, both of which are functions of the previous displacement estimate.
• memory 102 is also arranged to provide a set of appropriately chosen intensity values to interpolator 131, so as to generate a prediction p which will deviate from the present pel intensity value by as small an error as possible.
• interpolator 131 like interpolator 107, may be a conventional three or four point two-dimensional linear interpolator. The interpolator is made for the location in the previous frame which is identified by the displacement estimator, since the intensity value at that location should most closely match the intensity of the picture element presently being processed.
• the predicted intensity value output from interpolator 131 is applied to subtractor circuit 133 which forms an error signal which is the difference between the predicted and the actual intensity value of the input video signal on line 101. This difference is applied to a conventional quantizer 134.
• the error signal output from quantizer 134 is applied to a channel formatter 140 via a switch 136, the position of which is controlled by segmenter 109. With the switch closed, the quantized prediction error is encoded. However, when the switch is open, in picture areas that are called compensable or background, the value of is not transmitted to the receiver, since the value can be reconstructed adequately from previously encoded values.
• the predicted intensity value output from interpolator is combined with the error signal passed through switch 136 in an adder circuit 135 in order to generate a reconstructed or decoded intensity value
• This decoded value is re-entered in frame memory 102, via a switch 137, so that the intensity values stored in memory 102 are appropriately updated for picture elements in the moving area.
• switch 137 is arranged to recirculate the output of memory 102 to its input, on line 138.
• Switch 137 is arranged to operate in unison with switch 111 under control from segmenter 109.
• segmenter 109 is arranged to allow encoding of quantized prediction error values only within the moving area for non-compensable picture elements which, as will be defined below, are those pels for which the displaced frame difference DFD exceeds a threshold, indicating that motion compensated prediction is inadequate.
• switch 136 is closed, and the error value is coupled to formatter 140. With switch 136 open, and memory 102 receives either the predicted intensity value p (in moving areas) or the stored intensity value from the previous frame (in non-moving areas).
• segmenter 109 differentiates between moving and background areas by comparing a picture function (in this case frame difference (FDIF)) with a threshold.
• FDIF frame difference
• Segmenter 112 is shown in block di agram form in FIG . 2 .
• the s egmenter includes two comparators 201 and 202, each of which are arranged to compare an input signal to a preselected threshold value T 1 _ and T 2 , respectively.
• the input to comparator 201 is the intensity difference between each picture element and the spatially corresponding pel intensity value from the previous frame, as computed by subtractor circuit 210.
• a suitable frame delay may be provided by a separate delay circuit 205, or frame memory 102 can be used for this purpose.
• This intensity difference is, of course, the frame difference signal FDIF used in prior art conditional replenishment coders. If FDIF > T 1 , a. moving area (MA) is detected, raising the level on line 203. If the threshold is not exceeded, a non-moving or background area is present.
• the MA signal is used to control switches 111 and 137 of FIG. 1. Other more sophisticated techniques may be used for identifying moving areas in the picture,
• the input to comparator 202 is the unquantized motion compensated prediction error output from adder 133 of FIG. 1. If ⁇ > ⁇ 2 and, in addition, if a moving area has been detected, the output of AND gate 220, designated UC, goes high, indicating a non-compensable area for which a prediction error value must be transmitted. In this case, switch 136 connects the output of quantizer 134 to channel formatter 140. If the prediction error does not exceed the threshold, an error value is input to formatter 140 and to adder 135.
• channel formatter 140 of FIG. 1 receives the quantized prediction error ⁇ from quantizer 134 on line 301, when switch 136 is closed, and moving area (MA) and uncompensable area (UC) control signals on lines 203 and 204, respectively, from the segmenter of FIG. 2.
• the purpose of formatter is to encode address information indicating the different picture areas, i.e., (1) background, (2) compensable and (3) non-compensable areas, using run length encoding techniques, and to encode the prediction error values for the non-compensable pels, using variable length coding. Since three different types of areas are involved, a code indicating the type of picture area must also be transmitted. Since the statistical properties of run lengths of different types of area may vary, the formatter of FIG.
• 3 may include three separate code memories 304, 305 and 306, each of which stores a dictionary of run length code words used for a particular type of run: memory 304 stores run length codes for background areas, memory 305 stores run length codes for uncompensable areas, and memory 306 is used to store compensable area run length codes.
• the segment detector and encoder receives the MA and UC control signals and, in a manner to be explained below, generates a start of segment signal S on line 340. This signal is applied to run length encoder 308, which in turn, provides a run length to each of the code memories 304-306.
• a segment type signal on line 360 is also encoded and concurrently controls the position of a data selector 341, so that the run length code appropriate to the area type being processed is coupled back to encoder 308 on line 342.
• selector 341 assumes the first position for background pels, the second position for uncompensable pels, and the third position for compensable pels.
• the run length codes output from encoder 308 are stored in a first in/first out (FIFO) buffer 310 prior to being multiplexed into an output buffer 351.
• FIFO first in/first out
• variable length encoder 314 which may be a conventional coder.
• the variable length code words generated by coder 314 are entered in a second FIFO buffer 315, prior to entry in buffer 351, again via multiplexer 311. Since the code word T indicative of which area type is being encoded must itself be transmitted, the T code word on line 360 is applied to a segment type buffer 316, the output of which is also coupled to buffer 351 via multiplexer 311.
• the operating sequence of multiplexer 311 is controlled by a logic circuit 361, which is responsive to the segment type code word T generated by detector 320.
• logic 361 alternates between buffer 316 and buffer 310, such that a segment type code word is followed by a run length code indicating the number of consecutive picture elements which are of the same type, i.e., background or compensable moving area.
• the run length code for that area is followed by the variable length words stored in buffer 315 which indicate the prediction error values for each of that area's pels.
• FIG. 4 A block diagram of segment detector and encoder 320 is shown in FIG. 4.
• the moving area (MA) and uncompensable area (UC) signals on lines 203 and 204 are each applied to a respective flip-flop 402 and 403, both of which are reset at each pel time by a clock system pulse on line 401.
• the flip-flop outputs and inputs are then compared in a pair of exclusive OR gates 404 and 405. If either of the gate outputs is high, a segment transition has been detected, and the S signal is coupled to run length encoder 308 of FIG. 3 via an OR gate 406 and line 340.
• Detector 320 also includes a coder 410 which receives the MA and UC control signals and the S signal and generates an appropriate segment type code word T on line 360 which uniquely indicates the type of segment being processed. A code word is generated each time the S signal indicates the start of a new segment.
• the encoder of FIG. 1 is arranged, in accordance with the invention, to encode address information concerning three types of picture areas, namely background, compensable moving areas and non-compensable moving areas, and to encode prediction error values only for picture elements in the latter area. While displacement information is available in the encoder of FIG. 1, it is not encoded or transmitted, since accurate decoding can be accomplished, as described hereinafter. Should it be desired, however, to encode displacement data along with the addresses and prediction error values, this modification is readily made, as shown in FIG. 5. In this figure, elements identical to those of FIG. 1 retain the same reference designations.
• the displacement calculation in the encoder of FIG. 5 can be based on the actual input vi deo data , rather than on the reconstructed (decoded) intensity values output from adder circuit 135 or the recirculated background intensity values on line 138, as is the case in the apparatus of FIG. 1. Accordingly, line delay circuit 132 is eliminated, and the intensity values input to subtractor 103 are derived directly from input line 101.
• the estimates output from delay circuit 104 are applied to a register 501, which provides an output only when clock pulses are applied on line 503.
• the displacement estimate stored in register 501 is coupled to interpolator 131 and formatter 540.
• the displacement estimate is applied to frame memory 504 via quantizer 502, so that the frame memory can in turn apply appropriate intensity values of displaced picture elements to interpolator 131.
• a quantizer 105 continuously couples the output of delay circuit 104 to frame memory 504, so that the intensity values input to interpolator 107 reflect each recursively updated displacement estimate, rather than only those estimates clocked by pulses on line 503.
• the formatter 540 must be suitably modified to include the displacement estimates and unique flag words indicative thereof. This modification will be apparent to those skilled in the art. It is also to be understood that the displacement estimate values will themselves contain redundant information which can be decreased using established compression techniques.
• the picture intensity information encoded by the apparatus of FIG. 1 may be decoded with apparatus shown in block diagram form in FIG. 6.
• the coded input received on line 601 is applied to a deformatter 602, shown in detail in FIG. 7, which separates the incoming signal into error values on line 603 and an MA signal indicating moving area on line 604.
• the error values are applied to one input of an adder circuit 605, the other input of which receives predicted intensity values p from line 606.
• the output of adder circuit 605 is the decoded video signal, which is fed back to a frame memory 607 via a switch 608.
• the position of switch 608 is controlled by the MA signal on line 604, such that the decoded intensity value output from adder circuit 605 is entered in memory 607 in the moving area of the picture, but the previously stored intensity value is fed back on line 609 when MA is low, indicating a non-moving or background picture element.
• delay element 610 stores the previous displacement estimate which is updated by the output from displacement error calulator 611 in accordance with equation (1).
• the displacement estimate stored in delay circuit 610 is coupled to memory 607 via quantizer 618, which rounds or truncates the displacement value to enable selection of appropriate stored intensity values from the memory. These values, for picture elements in the neighborhood of the displaced location are applied to interpolator 613 on lines 615 together with the non-quantized value of on line 614. Interpolator 613, in turn, computes the intensity value at the displaced location and the intensity gradient at that location. The former is subtracted from a line delayed version of the present pel intensity value (via delay circuit 616) in a subtraction circuit 617, and the resulting displaced frame difference DFD is applied to displacement error calculator 611 together with the gradient value mentioned above.
• the output of quantizer 618 causes memory 607 to apply intensity values of pels which surround the displaced location to interpolator 619, which also receives the unquantized output of delay circuit 610.
• Interpolator 619 forms the predicted value p at the displaced location using conventional linear interpolation.
• the error values and the moving area signal (MA) are separated from the received coded input by deformatter 602, shown in detail in FIG. 7.
• the input data on line 601 is stored in a buffer 701, the output of which is fed through a data switch 702 to a segment detector and decoder 704.
• the outputs of detector and decoder 704 are first and second control signals on lines 705 and 706 which are used to recover the UC and MA control signals, respectively.
• the deformatter and decoder of FIG. 7 also includes logic 703 which causes switch 702 to alternate between detector/decoder 704 and a run length decoder 708, such that each segment type code is routed to the detector/decoder and each subsequent run length code is routed to decoder 708.
• logic 703 causes switch 702 to route the corresponding prediction error values to a variable length decoder 709 after run length code for that area has been applied to decoder 708.
• the run length code words applied to decoder 708 are routed to a series of code memories 710-712 which correspond to code memories 304-306 of FIG. 3. Specifically, memory 710 is arranged to decode runs in background areas, memory 711 is arranged to decode runs in uncompensable areas, and memory 712 is used for decoding compensable area runs. An output from the appropriate code memory is selected by the position of a data switch 713, which is controlled by the type code word T output from switch 702 on line 707.
• decoder 708 supplies a corresponding series of "1" bits to the clock inputs of first and second flip-flops 714 and 715.
• Flip-flop 714 is set by the first control signal output from detector/decoder 704 on line 706, and the output of this flip-flop is the MA control signal.
• Flip-flop 715 is set by the second control signal output from detector/decoder 704 on line 705, and the output of this flip-flop is the UC control signal.
• the prediction error code bits are decoded by decoder 709, which performs a function inverse to that of encoder 314 of FIG. 3.
• the output of decoder 709 which is the error value for each uncompensable picture element, is stored in a buffer 716, and connected to output line 603 when switch 717 is closed by the UC control signal output from flip-flop 715.
• Segment detector 704 may be arranged as shown in block diagram form in FIG. 8. Segment type code words T are received on line 707 and applied to a decoder 801 which is the inverse of encoder 410 of FIG. 4. Upon recognition of a code word on line 707, decoder 801 provides UC and MA control signal levels on lines 804 and 805, respectively, that correspond to the indicated segment type. These levels are clocked into respective flip-flops 802 and 803 when each word is decoded via a timing pulse on line 806. The outputs of flip-flops 802 and 803 are the first and second control signals referred to previously.

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Compression Or Coding Systems Of Tv Signals (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=21802172

Family Applications (1)

Country Status (6)

Cited By (2)

Families Citing this family (29)

Citations (6)

Family Cites Families (3)

• 1979
  • 1979-03-16 US US06/021,071 patent/US4218704A/en not_active Expired - Lifetime
• 1980
  • 1980-02-25 JP JP50064880A patent/JPS56500157A/ja active Pending
  • 1980-02-25 WO PCT/US1980/000177 patent/WO1980001976A1/en active Application Filing
  • 1980-02-25 GB GB8036286A patent/GB2060313B/en not_active Expired
  • 1980-02-25 DE DE3036769A patent/DE3036769C1/de not_active Expired
  • 1980-02-26 CA CA346,440A patent/CA1131349A/en not_active Expired
• 1982
  • 1982-03-09 CA CA000397975A patent/CA1137216A/en not_active Expired

Patent Citations (6)

Also Published As

Similar Documents

Legal Events