USRE34965E - Inter-frame predictive encoding system with encoded and transmitted prediction error - Google Patents

FIG. 1 shows a conventional inter-frame predictive encoding system. With reference to FIG. 1, a digitized video signal is fed to a

2 via an

1. The

2 subtracts a prediction digital signal from the digitized video signal to generate a prediction error (difference) signal. The prediction signal will be explained later.

The prediction error signal outputted from the

2 is subjected to an orthogonal transform by an

3. The orthogonal transform increases the efficiency of encoding. The output signal from the

3 is quantized by a

4. The output data from the

4 is encoded by an encoder S into variable-length codes such as Huffman codes. The codes outputted from the encoder S are transmitted via an

6.

The output data from the

4 is subjected to an inverse quantization by an

7 so that it is converted into a representative (setting of a representative). The representative outputted from the

7 is subjected to an inverse orthogonal transform by an

8. The output signal from the

8 corresponds to a prediction error (difference) which is generated through a decoding process in a decoding system explained later. An

9 adds the prediction error and a one-frame preceding prediction signal, generating a digitized video signal which corresponds to a digitized video signal generated by a decoding process in the decoding system. A

10 delays the output signal from the

9 by a time corresponding to a period of one frame. The output signal from the

10 is passed through a spatial low pass filter (a spatial LPF) 11, being converted into the prediction signal. The prediction signal is fed from the

11 to the

2 and the

9 via a

12. The

11 multiples the output data from the

9 by a coefficient which varies with a spatial frequency. The

11 ensures that a quantization error remains in the prediction error at a reduced rate. The

11 is effective and advantageous since a larger quantity of quantization errors are present in a high-frequency range and an inter-frame correlation is weakened by noises in the high-frequency range.

The

12 periodically couples and uncouples the

11 to and from the

2 and the

9 to periodically reset the inter-frame prediction. The period of the resetting the inter-frame prediction is generally set so as to correspond to 30 to 100 frames. When the inter-frame prediction is being reset, the prediction signal is fixed and essentially intra-frame coding is performed. The resetting prevents calculation errors from accumulating up to an unacceptable level. The calculation error would result from error codes generated in a transmission line and mismatching between orthogonal converters of a transmitter side and a receiver side in recursive-type inter-frame predictive en coding such as shown in FIG. 1. A shorter period of the resetting enables calculating errors from less accumulating but decreases the efficiency of the encoding.

FIG. 2 shows a conventional decoding system designed for the combination with the encoding system of FIG. 1. With reference to FIG. 2, variable-length digital data is fed to a

22 via an

21. The

22 converts the variable-length data into original fixed-length data. The output data from the

22 is subjected to an inverse quantization by an

23 so that it is converted into a representative (setting of a representative). The representative outputted from the

23 is subjected to an inverse orthogonal transform by an

24. The output signal from the

24 corresponds to a prediction error (difference). An

25 adds the prediction error and a one-frame preceding prediction signal, generating a digitized video signal. A

27 delays the output signal from the

25 by a time corresponding to a period of one frame. The output signal from the

27 is passed through a spatial low pass filter (a spatial LPF) 28, being converted into the prediction signal. The prediction signal is fed from the

28 to the

25. The

28 is similar to the

11 of the encoding system.

FIG. 4 shows an inter-frame predictive encoding system according to a first embodiment of this invention. With reference to FIG. 4, a digitized video signal is fed to a

37c of a

37 via an

1. The

37c of the

37 periodically changes between a first position and a second position at a predetermined period. When the

37c of the

37 assumes the first position, it contacts with a first

37a of the

37 and separates from a second

37b of the

37. When the

37c of the

37 assumes the second position, it contacts with the second

37b of the

37 and separates from the first

37a of the

37. As will be made clear hereinafter, when the

37c of the

37 contacts with the first

37a of the

37, the frame represented by the current input video signal is defined as an independent frame. Otherwise, the frame represented by the current input video signal is defined as a dependent frame.

When the

37c of the

37 contacts with the first

37a of the

37, the input video signal is transmitted through the

37 to a first

38a of a

38. A

38c of the

38 is connected to an input terminal of an

3. The

38c of the

38 periodically contacts with and separates from the first

38a and a second

38b of the

38 in a manner and at a timing similar to those of the

37. Specifically, when the

37c of the

37 contacts with the first

37a of the

37, the

38c of the

38 contacts with the first

38a of the

38 so that the input video signal is further transmitted through the

38 to the

3.

When the

37c of the

37 contacts with the second

37b of the

37 to a

31. The

31 delays the input video signal by a time corresponding to a predetermined number of frames. Specifically, the delay time of the

31 corresponds to (N-1) frames in the case where one independent frame occurs per N successive frames and the letter N denotes a predetermined natural number equal to two or greater. The output video signal from the

31 is fed to a

2. The

2 substracts a prediction signal from the video signal to generate a prediction error (difference) signal. The prediction signal will be explained later. The prediction error signal is outputted from the

2 to the second

38b of the

38. When the

37c of the

37 contacts with the second

37b of the

37, the

38c of the

38 contacts with the second

38b of the

38 so that the prediction error signal is transmitted through the

38 to the

3.

During one-frame periods separated by equal intervals corresponding to a predetermined number of frames, that is, during periods corresponding to independent frames, the

37c of the

37 contacts with the first

37a of the

37 while the

38c of the

38 contacts with the first

38a of the

38. During other periods, that is, during periods corresponding to dependent frames, the

37c of the

37 contacts with the second

37b of the

37 while the

38c of the

38 contacts with the second

38b of the

38.

The independent-frame video signal or the dependent-frame prediction error signal outputted from the

2 is subjected to an orthogonal transform by the

3. The orthogonal transform increases the efficiency of encoding. The output signal from the

3 is quantized by a

4. The output data from the

4 is encoded by an

5 into variable-length codes such as Huffman codes. The codes outputted from the

5 are transmitted via an

6.

The output data from the

4 is applied to a first

39a of a

39. A second fixed

39b of the

39 has no connection with other circuits. The

39c of the

39 is connected to an input terminal of an

7. The

39c of the

39 periodically contacts with and separates from the first

39a and a second

39b of the

39 in a manner and at a timing similar to those of the

37. Specifically, when the output signal from the

4 represents an independent frame, the

39c of the

39 contacts with the first

39a of the

39 so that the output signal from the

4 is transmitted to the

7. When the output signal from the

4 represents a dependent frame, the

39c of the

39 contacts with the second

39b of the

39 so that the transmission of the output signal from the

4 to the

7 is interrupted. In this way, only the output signal from the

4 which represents an independent frame is transmitted to the

7.

The independent-frame output signal from the

4 is subjected to an inverse quantization by the

7 so that it is convened into a representative (setting of a representative). The representative outputted from the

7 is subjected to an inverse orthogonal transform by an

8. The output signal from the

8 corresponds to a reproduced signal of an independent frame. The output signal from the

8, that is, the reproduced signal of an independent frame, is written into a

32.

A first fixed

40a of a

40 is connected to the output terminal of the

32. A second fixed

40b of the

40 has no connection with other circuits. A

40c of the

40 is connected to an input terminal of a

33. The

40c of the

40 periodically contacts with and separates from the first

40a and the second

40b of the

40 in a manner and at a timing similar to those of the

37. Specifically, when the output signal from the

8 which represents the current independent frame is written into the

32, the

40c of the

40 connects with the first

40a of the

40 so that the reproduced signal of the preceding independent frame is transferred from the

32 to the

33 via the

40. In this way, the reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame are prepared in the

32 and the

33 respectively.

The reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame remain stored in the

32 and the

33 respectively until the

32 is fed with the reproduced signal of the subsequent independent frame from the

8. The reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame are repeatedly outputted from the

32 and the

33 to

34 and 35 respectively. Specifically, the number of times of outputting the reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame is equal to N-1.

The

34 multiplies the reproduced signal of the current independent frame by a weight coefficient ct and outputs the resultant to an

36. The

35 multiplies the reproduced signal of the preceding independent frame by a weight coefficient (1-α) and outputs the resultant to the

36. The

36 adds the resultants of the multiplications, generating a prediction signal fed to the

2.

The weight coefficients a and (1-α) are determined in accordance with the time relation between the dependent frame inputted into the

2 and the independent frames related to the prediction signal inputted into the

2. For example, a linear prediction is used in the determination of the weight coefficients a and (1-α). Specifically, the weight coefficient α is given by the following equation.

In the encoding system of FIG. 4, the operation of the

37 and 38 determines independent frames which are separated in a time axis at equal intervals corresponding to a predetermined number of frames. In addition, dependent frames are defined between independent frames. The inter-frame correlation between coded data is cut at each independent frame. Therefore, a visual search is enabled by decoding only data of independent frames or by performing a random access in unit corresponding to independent frames.

In the encoding system of FIG. 4, the prediction signal is generated by adding the data of the two adjacent independent frames with variable weighting parameters. Specifically, the weight coefficients α and (1-α) used in the generation of the prediction signal are determined in accordance with the time relation between the dependent frame inputted into the

2 and the independent frames related to the prediction signal inputted into the

2. Therefore, the prediction can well follow the variation of display information between successive frames, and an S/N (a signal to noise ratio) of the prediction signal can be high.

The switches 37-40 are changed in response to a switch control signal. FIG. 11 shows a circuit for generating the switch control signal. As shown in FIG. 11, the switch control circuit includes a

501 which separates a frame sync signal from the input video signal. As shown in FIG. 12, the frame sync signal has a train of pulses. The pulses of the frame sync signal outputted from the

501 are counted by a

502. Each time four successive pulses of the frame sync signal are counted by the

502, the

502 outputs a pulse as shown in FIG. 12. The duration of each output pulse from the

502 agrees with the one-frame period. The sequentially-outputted pulses from the

502 compose the switch control signals fed to the switches 37-40.

FIG. 6 shows an inter-frame predictive decoding system according to the first embodiment of this invention. With reference to FIG. 6, variable-length digital data is fed to a

22 via an

21. The

22 converts the variable-length data into original fixed-length data. The output data from the

22 is subjected to an inverse quantization by an

23 so that it is converted into a representative (setting of a representative). The representative outputted from the

23 is subjected to an reverse orthogonal transform by an

24. For independent frames, the output signal from the

24 corresponds to a reproduced video signal. For dependent frames, the output signal from the

24 corresponds to a prediction error signal.

The output signal from the

24 is applied to a

47c of a

47. A first fixed

47a of the

47 is connected to an input terminal of a

42. A second fixed

47b of the

47 is connected to an

41. The

47c of the

47 periodically changes between a first position and a second position at a predetermined period. When the

47c of the

47 assumes the first position, it contacts with the first

47a of the

47 and separates from the second

47b of the

47. When the

47c of the

47 assumes the second position, it contacts with the second

47b of the

47 and separates from the first

47a of the

47. When the output signal from the

24 represents an independent frame, the

47c of the

47 contacts with the first

47a of the

47 so that the independent-frame signal is fed to and written into the

42. When the output signal from the

24 represents a dependent frame, the

47c of the

47 contacts with the second

47b of the

47 so that the dependent-frame signal is fed to the

41.

The

41 adds the dependent-frame signal and a prediction signal, reproducing a digitized video signal of a dependent frame. The prediction signal will be explained later. The

41 outputs the reproduced video signal of a dependent frame to a second fixed contact 48b of a switch 48. A first fixed

48a of the switch 48 is connected to a first

49a of a

49 which will be explained later. A

48c of the switch 48 is connected to an

26. The

48c of the switch 48 periodically contacts with and separates from the first

48a and the second fixed contact 48b of the switch 48 in a manner and at a timing similar to those of the

47. Specifically, when the

41 outputs the reproduced video signal of a dependent frame, the

48c of the switch 48 contacts with the second fixed contact 48b of the switch 48 so that the dependent-frame video signal is transmitted from the

41 to the

26 via the switch 48.

As described previously, the output signal from the

24 which agrees with the reproduced signal of an independent frame is written into the

42. A movable contact 49c of a

49 is connected to the output terminal of the

42. A first fixed

49a of the

49 is connected to an input terminal of a

43. A second fixed

49b of the

49 has no connection with other circuits. The movable contact 49c of the

49 periodically contacts with and separates from the first

49a and the second

49b of the

49 in a manner and at a timing similar to those of the

47. Specifically, when the output signal from the

24 represents the current independent frame and is thus written into the

42, the movable contact 49c of the

49 connects with the first

49a of the

49 so that the reproduced signal of the preceding independent frame is transferred from the

42 to the

43 via the

49. At the same time, the

48c of the switch 48 connects with the first

48a of the switch 48 so that the reproduced signal of the preceding independent frame is transferred from the

42 to the

26 via the

48 and 49. As understood from the previous description, the reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame are prepared in the

42 and the

43 respectively.

The reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame remain stored in the

42 and the

43 respectively until the

42 is fed with the reproduced signal of the subsequent independent frame from the

24. The reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame are repeatedly outputted from the

42 and the

43 to

44 and 45 respectively. Specifically, the number of times of outputting the reproduced signal of the current independent frame and the reproduced signal of the preceding independent frame is equal to N-1.

The multiplier 44 multiplies the reproduced signal of the current independent frame by a weight coefficient a and outputs the resultant to an adder 445. The

45 multiplies the reproduced signal of the current independent frame by a weight coefficient (1-α) and outputs the resultant to the adder 445. The

46 adds the resultants of the multiplications, generating a prediction signal fed to the

41. The weight coefficients α and (1-α) are determined similarly to the determination of the weight coefficients in the encoding system of FIG. 4.

During the processing of the input video signal by the encoding system of FIG. 4, independent frames are advanced relative to dependent frames. To compensate the advance of the independent frames, the reproduced video signal of an independent signal is outputted from the

42 to the

26 when the prediction process related to dependent frames between two independent frames is completed. Accordingly, the

42 also functions to perform the time correction.

In the encoding system of FIG. 7, the

7, the

8, and the switch 39 (see FIG. 4) are omitted while the input terminal of the

32 is connected to the first

37a of the

37.

When the input video signal of an independent frame is fed to the

3 via the

37 and 38, the independent-frame signal is also fed to the

32 and is written thereinto. Accordingly, the input video signal of independent frames is directly used in the generation of a prediction signal fed to the

2. It should be noted that, in the encoding system of FIG. 4, the resultant of the processing of the input video signal of an independent frame by the

3, 4, 7, and 8 is used in the generation of a prediction signal.

FIG. 8 shows an inter-frame predictive encoding system according to a third embodiment of this invention. With reference to FIG. 8, a digitized video signal is fed to a movable contact 142c of a

142 via an

101. The movable contact 142c of the

142 periodically changes between a first position and a second position at a predetermined period. When the movable contact 142c of the

142 assumes the first position, it contacts with a first fixed

142a of the

142 and separates from a second

142b of the

142. When the movable contact 142c of the

142 assumes the second position, it contacts with the second

142b of the

142 and separates from the first fixed

142a of the

142. As will be made clear hereinafter, when the movable contact 142c of the

142 contacts with the first fixed

142a of the

142, the frame represented by the current input video signal is defined as an independent frame. Otherwise, the frame represented by the current input video signal is defined as a dependent frame.

When the movable contact 142c of the

142 contacts with the first fixed

142a of the

142, the input video signal is transmitted through the

142 to a first fixed

143a of a

143. A movable contact 143c of the

143 is connected to an input terminal of an

103. The movable contact 143c of the

143 periodically contacts with and separates from the first fixed

143a and a second fixed contact 143b of the

143 in a manner and at a timing similar to those of the

142. Specifically, when the movable contact 142c of the

142 contacts with the first fixed

142a of the

142, the movable contact 143c of the

143 contacts with the first fixed

143a of the

143 so that the input video signal is further transmitted through the

143 to the

103.

When the movable contact 142c of the

142 contacts with the second

142b of the

142, the input video signal is transmitted through the

142 to a

131. The

131 delays the input video signal by a time corresponding to a predetermined number of frames. Specifically, the delay time of the

131 corresponds to (N-1) frames in the case where one independent frame occurs per N successive frames and the letter N denotes a predetermined natural number equal to two or greater. The output video signal from the

1;31 is fed to a

102. The

102 subtracts a prediction signal from the video signal to generate a prediction error (difference) signal. The prediction signal will be explained later. The prediction error signal is outputted from the

102 to the second fixed contact 143b of the

143. When the movable contact 142c of the

142 contacts with the second

142b of the

142, the movable contact 143c of the

143 contacts with the second fixed contact 143b of the

143 so that the prediction error signal is transmitted through the

143 to the

103.

During one-frame periods separated by equal intervals corresponding to a predetermined number of frames, that is, during periods corresponding to independent frames, the movable contact 142c of the

142 contacts with the first fixed

142a of the

142 contacts while the movable contact 143c of the

143 contacts with the first fixed

143a of the

143. During other periods, that is, during periods corresponding to dependent frames, the movable contact 142c of the

142 contacts with the second

142b of the

142 while the movable contact 143c of the

143 contacts with the second fixed contact 143b of the

143.

The video signal of the independent-frame or the prediction error signal of the dependent-frame outputted from the

102 is subjected to an orthogonal transform by the

103. The orthogonal transform increases the efficiency of encoding. The output signal from the

103 is quantized by a

104. The output data from the

104 is encoded by an

105 into variable-length codes such as Huffman codes. The codes outputted from the

105 are transmitted via an

106.

The independent-frame video signal is transmitted via the

142 and is written into a

132. A movable contact 144c of a

144 is connected to an output terminal of the

132. A first fixed

144a of the

144 is connected to an input terminal of a

133. A second fixed

144b of the

144 has no connection with other circuits. The movable contact 144c of the

144 periodically contacts with and separates from the first fixed

144a and the second

144b of the

144 in a manner and at a timing similar to those of the

142. Specifically, when the current independent-frame signal is transmitted via the

142 and is written into the

132, the movable contact 144e of the

144 connects with the first fixed

144a of the

144 so that the preceding independent-frame signal is transferred from the

132 to the

133 via the

144. In this way, the current independent-frame signal and the preceding independent-frame signal are prepared in the

132 and the

133 respectively.

The current independent-frame signal and the preceding independent-frame signal remain stored in the

132 and the

133 respectively until the

132 is fed with the subsequent independent-frame signal via the

142. The current independent-frame signal and the preceding independent-frame signal are repeatedly outputted from the

132 and the

133 to position

134 and 135 respectively. The

134 two-dimensionally shifts the current independent-frame signal by a magnitude which is determined by a

136. Similarly, the

135 shifts the preceding independent-frame signal by a magnitude which is determined by the

136.

The

134 includes an address generator, and a temporal memory into and from which the current independent-frame signal is written and read in accordance with an address signal from the address generator. During the reading out the signal from the temporal memory, the address generator shifts addresses relative to the write addresses in accordance with a signal from the

136 to provide the two dimensional shift of the current independent-frame signal. The

135 is designed similarly to the

134.

An output signal from a

141 which represents a motion vector is inputted into the

136. The

136 multiplies the motion vector by (N-i), generating a magnitude of shift fed to the

134. In addition, the

136 multiplies the motion vector by (-i), generating a magnitude of shift fed to the

135. The character i denotes the order number of a predicted frame which is determined in view of the time relation between frames. Specifically, the number i is 0 for each independent frame and varies as i=1, 2, 3, . . . , (N-i) for dependent frames.

Output signals from the

134 and 135 are fed to

137 and 138 respectively. The

137 multiplies the output signal from the

137 by a weight coefficient a and feeds the resultant to an

139. The

138 multiplies the output signal from the

135 by a weight coefficient (1-α) and feeds the resultant to the

139. The

139 adds the resultants of the multiplications, generating a prediction signal fed to the

102. For example, according to a linear prediction, the weight coefficient α is defined as i/N.

The current independent-frame video signal is transmitted via the

142 to a

140. The preceding independent-frame video signal is fed from the

132 to the

140. The

140 detects motion vectors from the current independent-frame signal and the preceding independent-frame signal in a known way such as a block matching method. The

140 outputs a signal representative of the detected motion vectors to the

141.

As shown in FIG. 9, the

140 includes

153 and 154 receiving the current independent-frame signal and the preceding independent frame signal via

151 and 152 respectively. The current independent-frame signal and the preceding independent-frame signal are written into the

153 and 154 respectively. Each of the signals stored in the

153 and 154 is divided into spatial regions whose number is greater than the number of spatial regions corresponding to blocks of 8×8 or 16×16 pixels which are used in determining motion vectors.

During the writing of the signal into each of the

153 and 154, addresses fed to each of the

53 and 154 are directly made by main addresses of pixels in blocks. During the reading of the signal from each of the

153 and 154, addresses fed to each of the

153 and 154 are modified by a value of motion vectors which are generated in a vector generator 55. The writing of the signal into each of the

53 and 154 is performed once for each independent frame, while the reading of the signal from each of the

153 and 154 is repeatedly performed a number of times which equals the number of motion vectors. The address shift is composed of a vector which is generated by a

155 in correspondence with intervals of N frames. For example, the

155 includes a counter.

Specifically, addresses to the

153 for the current independent frame are generated by adding the vector output from the

155 to an output signal from a

156. This addition is performed by an

157. For example, the

156 includes a counter. Addresses to the

154 for the preceding independent frame are generated by subtracting the vector output of the

155 from an output signal of the

156. This subtraction is performed by a

158.

A

159 calculates a difference between the output data from the

153 and 154. A squaring

160 calculates the square of the data difference. An

161 accumulates the output data from the squaring

160 during an interval corresponding to a period of one block. The

161 obtains a mean square error for a setting vector.

The section corresponding to a part surrounded by the broken line of FIG. 9 includes the

153, 154, 156, 157, 158, 159, 160, and 161. This section performs a process of detecting a mean square error. For each setting vector in each block, the detecting process is repeatedly performed a number of times which equals the number of pixels within one block.

The mean square errors of respective vectors are sequentially fed from the

161 to a

162. The

162 detects the smallest among the mean square errors and outputs the motion vector corresponding to the selected smallest mean square error as a motion vector V' representing a motion corresponding to each of blocks of one frame. For example, the

162 includes a smallest value selector of the serial input type which is composed of a combination of latches and comparators. The; motion vector V' is generated for each block. The motion vectors of one frame are stored into a

163. During the prediction process for dependent frames between two adjacent independent frames, the motion vectors are repeatedly read out from the

163 and are transmitted via an

164 to the

141.

Returning to FIG. 8, the

141 performs a process of calculating differences between vectors in each predicted frame. As will be made clear hereinafter, some circuit components are common to the

141 and the

140.

The

141 is fed with the dependent-frame signal from the

131. A first fixed

145d and a second fixed

145e of a

145 are connected to the output terminals of the

132 and 133 respectively. A

145f of the

145 is connected to the

141. The

145f of the

145 periodically changes between a first position and a second position at a predetermined period. When the

145f of the

145 assumes the first position, it contacts with the first fixed

145d of the

145 and separates from the second fixed

145e of the

145 so that the preceding independent-frame signal is fed from the

133 to the

141 via the

145. When the

145f of the

145 assumes the second position, it contacts with the second fixed

145e of the

145 and separates from the first fixed

145d of the

145 so that the current independent-frame signal is fed from the

132 to the

141 via the

145. In this way, the current independent-frame signal and the preceding independent-frame signal are alternately fed to the

141.

Specifically, the

145 is changed in response to the order number of a frame so that one of the current independent-frame signal and the preceding independent-frame signal is alternately and periodically selected and fed to the

141. Selected one of the current independent-frame signal and the preceding independent-frame signal relates to a larger weight in the predictive calculation.

As shown in FIG. 10, the

141 includes the

153 and 154 receiving the dependent-frame signal and the independent-frame signal via the

151 and 152 respectively. The dependent-frame signal and the independent-frame signal are written into the

153 and 154 respectively.

The motion vector V' determined between the current independent frame and the preceding independent frame is transmitted from the

140 to an

167 via an

165. The

167 adds the motion vector V' and an output value vd from a

166, generating a resultant vector. For example, the

166 includes a counter.

A known detection of motion vectors use a method in which first vectors are set at a low distribution density and then second vectors are set around the first vectors at a high distribution density. The output value vd from the

166 corresponds to a vector difference determined in respect of setting the second vectors in the known method.

A

168 multiplies the resultant vector by (-i) for the preceding independent frame. The

168 multiplies the resultant vector by (N-i) for the current independent frame. The character i denotes the order number of a predicted frame which is determined in view of the time relation between frames. Specifically, the number i is 0 for each independent frame and varies as i=1, 2, 3, . . . , (N-i) for dependent frames. The output signal from the

168 represents an address shift.

During the writing of the signal into each of the

153 and 154, addresses fed to each of the

153 and 154 are directly made by main addresses of pixels in blocks. During the reading of the signal from the

153, addresses fed to the

153 are also directly made by the main addresses. During the reading of the signal from the

154, addresses fed to the

154 are made by adding an address shift to the main addresses.

Specifically, reading addresses to the

154 for the independent frame are generated by adding the address shift from the

168 to the output signal from the

156. This addition is performed by an

169. The output signal from the

156 are fed to the

153 as reading and writing addresses.

The

159 calculates a difference between the output data from the

153 and 154. The squaring

160 calculates the square of the data difference. The

161 accumulates the output data from the squaring

160 during an interval corresponding to a period of one block. The

161 obtains a mean square error for a setting vector.

The mean square errors of respective vectors are sequentially fed from the

161 to the

162. The

162 detects the smallest among the mean square errors and outputs the motion vector corresponding to the detected smallest mean square error as a motion vector dV' representing a motion corresponding to each of blocks of one frame. The motion vector dV' is generated for each block. An

170 adds the motion vector dV' and the motion vector V', generating a final motion vector V. The final motion vector V outputted from the

170 is transmitted via the

164 to the

136 and an encoder 113 (see FIG. 8). The output data from the

141 is encoded by the

113 into variable-length codes such as Huffman codes.

The processing by the

140 and the processing by the

141 are performed during different periods respectively. This design enables reliable operation of the

140 and 141 although the circuit components are common to the

140 and 141. It should be noted that the circuit components of the

140 may separate from the circuit components of the

141.