Pulse video data lossless coding method based on decision tree prediction model

Abstract

The invention discloses a pulse video data lossless coding method based on a decision tree prediction model, and belongs to the technical field of high-speed photography. According to the method, feature extraction is carried out on each pixel based on a historical microstring, prediction is carried out through a decision tree prediction model, and a sparse 0 / 1 prediction label judgment sequence is generated by comparing a pixel value of a subsequent frame with a prediction value, so that the number of '1' in each frame of data is effectively reduced, and high-efficiency compression is realized. Meanwhile, LWZ data compression coding and 0-order exponential Golomb coding secondary compression are adopted for a prediction label judgment sequence, and a higher compression ratio is achieved through cascade coding.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of high-speed photography, in particular to a pulse video data lossless encoding method based on a decision tree prediction model. BACKGROUND

[0002] With the continuous development of video technology, people's demand for shooting high-speed moving objects is also becoming more and more urgent, and a new type of high-speed photography technology called 'pulse video' has emerged. The shooting principle of this technology is different from that of traditional technology. It adopts an integration comparison mechanism. The integrator can accumulate the charge output by the photoelectric converter, and the comparator compares the charge on the integrator with the preset threshold at fixed time intervals. If the charge accumulated on the integrator reaches the preset threshold, 1 is output and the charge on the integrator is emptied. If the charge on the integrator does not reach the preset threshold, 0 is output and the charge on the integrator is maintained. As can be seen, this method is very different from the traditional shooting method. The traditional shooting method is to extract the charge on the integrator at fixed time intervals, regardless of the amount of charge.

[0003] Pulse video better overcomes two problems of traditional high-speed photography. First, pulse video does not need to extract the specific intensity of the signal at each time interval, but only needs to compare the signal intensity (charge number) with the preset threshold. Second, only 1 bit of data is generated for each pixel at each time interval, reducing the data volume.

[0004] Although the data volume of pulse video is reduced, the sampling frequency is very high, and even if each pixel samples only 1 bit at each time, the data rate is still very high, which requires compression and encoding of pulse video data. However, the existing technology, such as the ordinary lossless file compression tool, such as winzip and winrar, still has a large space for improvement in the lossless compression efficiency of pulse video data. SUMMARY

[0005] In view of the defects in the prior art, the present application trains a decision tree model by subjecting pulse raw data to corresponding feature extraction processing to meet the specification of decision tree input data, thereby predicting the data at the next time and outputting a "predicted label judgment value". By using the decision tree plus LZW method, data redundancy can be greatly reduced and compression efficiency can be improved.

[0006] In order to achieve the above purpose, the present application provides a pulse video data lossless encoding method based on a decision tree prediction model, comprising: (1) obtaining a segment of original pulse video data with N frames, wherein each frame of the original pulse video data includes H rows of pixels, and each row has W pixels; (2) opening storage space, including: opening two first cache area P[k][i] and second cache area M[i][k] with a capacity of W*H*N, for caching label judgment data and pixel bit string data, i =0,1,2,...,W*H-1, k=0,…,N-1; (3) starting from k=1 until k=N, processing original pulse video data frame by frame, including: (3.1) when k<=3, the label judgment data is the original pulse video data, scanning each frame of original pulse video data according to the preset scanning mode, and storing in P[k-1][i] in time order; (3.2) when K=3, the pixel data of the same pixel position in the previous 3 frames is arranged in time order to obtain the pixel bit string of the pixel position, and stored in M[i][j] in time order, i =0,1,2,...,W*H-1, j=0,1,2;Each of the pixel bit string is divided into several micro strings, the micro string is a 1 and several 0s in front of it in the pixel bit string, until the front of a 1 or no pixel; (3.3) when k>3, for each pixel in the frame, judge the number n i of micro strings contained in the pixel bit string corresponding to the pixel position up to the previous frame i ; If n i >=3, feature extraction is performed on the pixel position, and a trained decision tree prediction model is called according to the extracted features to predict the pixel, if the predicted value is the same as the actual value R i , then P[k-1][i]=0 is obtained, otherwise P[k-1][i]==1 is obtained; If n i <3, no processing is done on the actual value R i , and P[k-1][i]=R i is directly obtained. The actual value R i is added to the corresponding position in the pixel bit string, M[i][k-1]= R i . (3.4) k=k+1, repeat steps (3.3)-(3.4) until k=N; (4) encoding the first cache area P[k][i] frame by frame according to the preset encoding method, and generating a code stream in time order.

[0007] Further, the feature extraction in step (3.3) is specifically: including 5 features: Feature one is the length of the nearest complete microstring to the pixel position, denoted as feat1; Feature two is the number of 0s accumulated before the pixel position in the microstring, denoted as feat2; Feature three is feat2 minus feat1; Feature four is the length difference between the two microstrings before the pixel position; Feature five is the length difference between the second microstring and the third microstring before the pixel position.

[0008] Further, the preset scanning mode is raster scanning.

[0009] Further, the prediction model of the decision tree is determined by the number of feature extraction, the number of layers of the tree, and the complete split method is used for tree splitting.

[0010] Further, the preset encoding method is to sequentially perform LWZ data compression encoding and 0-order exponential Golomb encoding.

[0011] Further, the LWZ data compression encoding is specifically: (4.1) Initialize the code table: encode 257 0s as 1, encode "1" as 2, encode "01" as 3, and so on, encode 256 0s and 1 1 as 258; when a new combination mode is encountered, the code value is accumulated from 259 downwards; (4.2) Perform LWZ encoding on the first cache area P[k][i] obtained in step (3) to form X k data, if the last bit of the current frame data is not 1, then a 0 is appended to the X k data.

[0012] The application also provides a pulse video data lossless decoding method based on a decision tree prediction model, comprising the following steps: (1) Obtain pulse video encoding data, and the frame number N, the number of pixel rows H of each frame, and the number of pixels W of each row of the pulse video encoding data; (2) Decode the pulse video encoding data according to a preset encoding method to obtain label judgment data; (3) Open a storage space, including opening two third cache areas O[k][i] and fourth cache areas N[i][k] with a capacity of W*H*N, which are used to cache original pulse video data and pixel bit string data, i = 0, 1, 2,..., W*H-1, k = 0,..., N-1; (3) Start from k = 1 and end at k = N, process the label judgment data frame by frame, including: (3.1) When k<=3, the label judgment data is the original pulse video data, the label judgment data is read by frame, and is sequentially stored in O[k-1][i] in time order, i=0, 1, 2,..., W*H-1; (3.2) When k=3, the pixel data of the same pixel position in the first 3 frames of the label judgment data is arranged in time order to obtain a pixel bit string of the pixel position, and is stored in N[i][j], i=0, 1, 2,..., W*H-1, j=0, 1, 2; each pixel bit string is divided into a plurality of micro-strings, the micro-string being a 1 and a plurality of 0s in front of it in the pixel bit string until a 1 in front or no pixel in front; (3.3) When k>3, for each pixel position in the frame, determine the number n of micro-strings contained in the pixel bit string corresponding to the pixel position up to the previous frame i , i represents the pixel position; If n i >=3, feature extraction is performed on the pixel position, and a trained decision tree prediction model is called according to the extracted features to predict the pixel to obtain a prediction value Pre, if the label judgment data corresponding to the pixel position is 0, then O[k-1][i]=Pre, otherwise O[k-1][i]=-Pre; If n i <3, the label judgment data is the original pulse video data, and the label judgment data of the pixel position in the current frame is directly stored in O[k-1][i]; The original pulse video data O[k-1][i] is added to the corresponding position in the pixel bit string; (3.4) k=k+1, repeat steps (3.3)-(3.4) until k=N.

[0013] Further, the feature extraction in step (3.3) is specifically: including five features: Feature one is the length of the nearest complete micro-string to the pixel position, denoted as feat1; Feature two is the number of 0s accumulated before the micro-string in which the pixel position is located, denoted as feat2; Feature three is feat2 minus feat1; Feature four is the length difference of the two micro-strings before the pixel position; Feature five is the length difference between the second micro-string and the third micro-string before the pixel position.

[0014] Further, the preset encoding method is to sequentially perform 0-order exponential Golomb decoding and LWZ data compression decoding.

[0015] Furthermore, the LWZ data compression encoding specifically includes: (2.1) Initialize the code table: Encode 257 zeros as 1, "1" as 2, "01" as 3, and so on. 256 zeros plus 1 1 are encoded as 258. When a new combination is encountered, the code value is accumulated from 259 downwards. (2.2) Obtain the length len of each frame from the decoded data of the 0th order exponent Columbus. i Frame by frame, LWZ decoding is performed by referring to the code table. If the last bit of the current frame data is 0, then (W*H - len) is added. i Add zeros to the current frame so that the data length of the current frame is W*H, where i represents the current frame number.

[0016] The beneficial effects of this invention are: 1. This invention extracts features from the historical micro-strings of each pixel, makes predictions using a decision tree prediction model, and compares the pixel values ​​of subsequent frames with the predicted values ​​to generate a sparse 0 / 1 prediction label judgment sequence, which effectively reduces the number of "1"s in each frame of data and achieves high-efficiency compression.

[0017] 2. This invention employs LWZ data compression encoding and 0th-order exponential Golomb encoding for secondary compression of the predicted label judgment sequence, and cascade encoding to achieve a higher compression ratio.

[0018] 3. During decoding, this invention reverse reuses the decision tree prediction model to accurately restore the original data from the label judgment value and the prediction value, while ensuring pixel-level lossless recovery. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the lossless encoding method for pulse video data based on a decision tree prediction model, as shown in an embodiment of the present invention.

[0020] Figure 2 This is the spatiotemporal structure of pulse video data in an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the method for dividing a pixel bit string into microstrings according to an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments. It should be noted that the embodiments of the present invention describe the process steps in a specific order; however, this is only for illustrating the specific embodiment and not for limiting the order of the steps. Conversely, in different embodiments of the present invention, the order of the steps can be adjusted according to process modifications.

[0023] like Figure 1As shown, the present application provides a pulse video data lossless encoding method based on decision tree prediction model, comprising: S101, obtaining a piece of original pulse video data with frame number N, wherein each frame of the original pulse video data comprises H rows of pixels, and the number of pixels in each row is W.

[0024] As shown, Figure 2 The video data has a data matrix with size W*H at each time, which is called "frame". Each position of the matrix corresponds to a pixel. The data in each frame of the traditional video data is an integer of 8 or more bits.

[0025] S102, opening a storage space, comprising: opening two first buffer areas P[k][i] and second buffer areas M[i][k] with capacity W*H*N, for buffering label judgment data and pixel bit string data.

[0026] S103, starting from k=1 to k=N, processing the original pulse video data frame by frame, comprising: (1) When k<=3, the label judgment data is the original pulse video data, each frame of the original pulse video data is scanned according to the preset scanning mode, and is sequentially stored in P[k-1][i] according to time sequence.

[0027] Among them, the preset scanning mode is raster scanning.

[0028] (2) When K=3, the pixel data of the same pixel position in the previous 3 frames is arranged into a string in time sequence, to obtain a pixel bit string of the pixel position, and is stored in M[i][j] in time sequence, i=0,1,2,...,W*H-1,j=0,1,2; each pixel bit string is divided into several micro strings, which are a 1 in the pixel bit string and several 0s in front of it until a 1 in front or no pixel in front.

[0029] The data of the same position in the data matrix at different times is extracted and arranged in time sequence to form the data of the pixel, also called pixel bit string. As shown, Figure 3 The pixel bit string in the figure contains 13 time sampling data, that is, 13 bits. Among them, there are 4 1s. According to the method shown in the figure, it is divided into 4 micro strings.

[0030] (3) When k>3, for each pixel in the frame, judge the number n i of micro strings contained in the pixel bit string corresponding to the pixel position of the previous frame, and obtain the actual value R i of the position in the current frame according to the preset scanning mode.

[0031] If n i>=3, the pixel position is extracted and the trained decision tree prediction model is called to predict the pixel according to the extracted features. If the predicted value is equal to the actual value R i , P[k-1][i]=0, otherwise P[k-1][i]=1.

[0032] If n i <3, the actual value R i is not processed, and P[k-1][i]=R i is directly obtained.

[0033] The actual value R i is added to the corresponding position in the pixel bit string, and M[i][k-1]=R i .

[0034] Among them, the extracted features include five features: Feature one is the length of the nearest complete microstring to the pixel position, denoted as feat1.

[0035] Feature two is the number of 0s accumulated before the pixel position in the microstring, denoted as feat2.

[0036] Feature three is feat2 minus feat1.

[0037] Feature four is the length difference of the previous two microstrings of the pixel position.

[0038] Feature five is the length difference between the second and third microstrings of the pixel position.

[0039] It can be seen here that in order to predict, the number of microstrings composed of data at the previous time of the pixel position to be predicted must be greater than or equal to 3, so that the corresponding feature extraction can be completely performed. Therefore, the pixel data contained in the first three microstrings of each pixel is not predicted and the original value of the data is directly stored.

[0040] The prediction model of the decision tree uses the most traditional decision tree algorithm to determine the number of layers of the tree according to the number of features. Here, the number of features is 5, so the maximum depth of the tree is also 5. Since the feature values are discrete and the number of categories is limited to a short range, a complete splitting method is used for the splitting of the tree here, that is, each feature item has a corresponding feature value, and the number of branches of each node is equal to the number of feature values. Entropy is used to represent a measure of the degree of data order. First, the empirical entropy is calculated according to the label data, denoted by H(D). Then, the conditional entropy H(D|A) of the data D divided by each feature A is calculated. Finally, the information gain (IG(D,A)) is used to determine which feature to select as the division node. The larger the IG(D,A), the better the division effect (IG(D,A)=H(D)-H(D|A)). Recursively build the tree until the termination condition is met. The termination conditions are as follows: the data purity of the node reaches 100%; the maximum tree depth is reached; and IG(D,A) of all features <=0. When the termination condition is met, a leaf node is generated, which no longer continues to split, but stores the class probability P(class=1). When P>0.5, it is judged as class 1.

[0041] (4) k=k+1, repeat steps (3)-(4) until k=N.

[0042] The following illustrates micro strings and the prediction process: For example: 001 0001 00001 00001 000001 (this is a bit stream of a pixel) 234 00001 000001 (this is the generated 3 micro strings and the bit stream to be predicted) Start extracting features and prediction: pre 00011 000001 (generated prediction value) 001 0001 00001 00010 000000 (code stream saved after label judgment) Note: Assume that this data is the data at the beginning of the file.

[0043] S104, encode the first cache area P[k][i] frame by frame according to the preset encoding method, and sort according to time to generate a code stream.

[0044] The preset encoding method is to sequentially perform LWZ data compression encoding and 0-order exponential Golomb encoding.

[0045] Use the dictionary-based compression method (LZW) for preliminary compression: read each frame of W*H pixel bits in raster scan order, and then use the LZW method for frame-by-frame compression. LZW encoding scheme design: (1) initialization code table: 257 0s are encoded as 1, 1 is encoded as 2, 01 is encoded as 3, and so on, and 256 0s and 1 1 are encoded as 258; when a new combination mode is encountered, the code value is accumulated from 259 downwards.

[0046] (2) the first cache area P[k][i] obtained in step S103 is subjected to LWZ encoding frame by frame, and each frame of data continues to use the code table data cached in the last frame to encode, until the entire pulse label judgment file scanning ends. After each frame of data is subjected to LZW encoding, X k coding data is formed, and if the last bit of the current frame of data is not 1, a 0 is supplemented after the X k data.

[0047] After the data subjected to LZW preliminary compression, the data is subjected to 0-order exponential Golomb encoding, and the result is sorted according to time to form the coding data of the pulse video data.

[0048] The application also provides a pulse video data lossless decoding method based on a decision tree prediction model, comprising the following steps: S201, acquiring pulse video coding data, and the frame number N, the number of pixel rows H of each frame, and the number of pixels W of each row of the pulse video coding data.

[0049] S202, decoding the pulse video coding data according to a preset encoding method to obtain label judgment data.

[0050] Corresponding to the encoding method, first, 0-order exponential Golomb decoding is performed frame by frame, and then LWZ decoding is performed.

[0051] During LWZ decoding, the initialization code table is consistent with the encoding table, and here, the length len i of each frame is acquired in advance, the data is read frame by frame, the code table is compared, and LWZ decoding is performed, if the last bit of the current frame of data is 0, (W*H - len i ) 0s are supplemented to the current frame, so that the length of the current frame of data is W*H. Then, the decoding of the next frame is performed, and i represents the current frame number.

[0052] S203, opening a storage space, comprising: opening two third cache areas O[k][i] and fourth cache areas N[i][k] with a capacity of W*H*N, for caching original pulse video data and pixel bit string data.

[0053] S204, starting from k=1 to k=N, frame by frame processing label judgment data, comprising: (1) When k<=3, the label judgment data is the original pulse video data, the label judgment data is read by frame, and is sequentially stored in O[k-1][i] according to time sequence, i=0, 1, 2,..., W*H-1.

[0054] (2) When K=3, the pixel data of the same pixel position in the first 3 frames of the label judgment data is arranged into a string according to time sequence, to obtain a pixel bit string of the pixel position, which is stored in N[i][j], i=0, 1, 2,..., W*H-1, j=0, 1, 2; each pixel bit string is divided into a plurality of microstrings, the microstring being a 1 in the pixel bit string and a plurality of 0s in front of the 1 until a 1 in front or no pixel in front.

[0055] (3) When k>3, for each pixel position in the frame, the number n of microstrings contained in the pixel bit string corresponding to the pixel position in the previous frame is determined. i , i represents the pixel position.

[0056] If n i >=3, the feature extraction is performed on the pixel position, and the trained decision tree prediction model is called according to the extracted feature to predict the pixel, to obtain a prediction value Pre, if the label judgment data corresponding to the pixel position is 0, then O[k-1][i]=Pre, otherwise O[k-1][i]=-Pre.

[0057] If n i <3, the label judgment data is the original pulse video data, and the label judgment data of the pixel position in the current frame is directly stored in O[k-1][i].

[0058] The original pulse video data O[k][i] is added to the corresponding position in the pixel bit string.

[0059] The feature extraction and the decision tree prediction model are described in the encoding method.

[0060] (4) k=k+1, steps (3)-(4) are repeated until k=N.

[0061] Although the above describes the embodiments of the present application, it should be understood that they are presented only as examples, not as limitations. It is obvious to those skilled in the relevant art that various combinations, variations and changes can be made without departing from the spirit and scope of the present application. Therefore, the width and scope of the present application disclosed herein should not be limited by the above disclosed exemplary embodiments, but should be defined only according to the appended claims and their equivalent replacements.

Claims

1. A lossless coding method for pulse video data based on a decision tree prediction model, characterized in that, Includes the following steps: (1) Obtain a raw pulse video data with a frame number of N, wherein each frame of the raw pulse video data includes H rows of pixels and the number of pixels in each row is W; (2) Allocate storage space, including: allocating two buffers, a first buffer P[k][i] and a second buffer M[i][k], with a capacity of W*H*N, to cache tag judgment data and pixel bit string data, i =0,1,2,...,W*H-1, k=0,…,N-1; (3) Process the raw pulse video data frame by frame from k=1 to k=N, including: (3.1) When k<=3, the tag judgment data is the original pulse video data. Scan each frame of original pulse video data according to the preset scanning method, and store it into P[k-1][i] in order of time. (3.2) When K=3, arrange the pixel data of the same pixel position in the first 3 frames into a string according to the time order to obtain the pixel bit string of that pixel position, and store it into M[i][j] according to the time order, i =0,1,2,...,W*H-1, j=0,1,2; divide each pixel bit string into several micro strings, the micro string is a 1 in the pixel bit string and several 0s before it, until the preceding 1 or no preceding pixel; (3.3) When k>3, for each pixel in the frame, determine the number n of microstrings in the pixel bit string corresponding to the pixel position in the previous frame. i And obtain the actual value R at that position in the current frame according to the preset scanning method. i ; If n i If the value is greater than or equal to 3, then feature extraction is performed on the pixel location, and the trained decision tree prediction model is called to predict the pixel based on the extracted features. If the predicted value differs from the actual value R... i If they are the same, then P[k-1][i]=0; otherwise, P[k-1][i]==1. If n i <3, not the actual value R i After processing, we can directly obtain P[k-1][i]=R i ; The actual value R i The corresponding position in the added pixel bit string, M[i][k-1] = R i ; (3.4) k = k + 1, repeat steps (3.3) - (3.4) until k = N; (4) Encode the first buffer P[k][i] frame by frame according to the preset encoding method, and sort them by time to generate the bit stream.

2. The lossless encoding method for pulse video data based on a decision tree prediction model according to claim 1, characterized in that, The feature extraction in step (3.3) specifically involves: Includes 5 features: feature First, the length of the nearest complete microstring to this pixel location, denoted as feat1; Feature 2 is the number of previously accumulated 0s in the microstring containing the pixel position, denoted as feat2; Feature 3: feat2 minus feat1; Feature 4 is the difference in length between the two microstrings preceding the pixel position; Feature 5 is the length difference between the second and third microstrings preceding the pixel position.

3. The lossless encoding method for pulse video data based on a decision tree prediction model according to claim 1, characterized in that: The preset scanning method is raster scanning.

4. The lossless encoding method for pulse video data based on a decision tree prediction model according to claim 1, characterized in that: The prediction model of the decision tree determines the number of tree layers based on the number of features extracted, and the tree splits using the full split method.

5. The lossless encoding method for pulse video data based on a decision tree prediction model according to claim 1, characterized in that: The preset encoding method is to sequentially perform LWZ data compression encoding and 0th-order exponential Golomb encoding.

6. The lossless coding method for pulse video data based on a decision tree prediction model according to claim 5, characterized in that, The LWZ data compression encoding is specifically as follows: (4.1) Initialize the code table: Encode 257 zeros as 1, "1" as 2, "01" as 3, and so on. Encode 256 zeros and 1 one as 258. When a new combination is encountered, the code value is accumulated from 259 downwards. (4.2) The first buffer P[k][i] obtained in step (3) is LWZ encoded frame by frame to form X. k If the last bit of the current frame data is not 1, then in X... k Add a 0 after each data point.

7. A lossless decoding method for pulse video data based on a decision tree prediction model, characterized in that, Includes the following steps: (1) Obtain pulse video encoded data, and the number of frames N, the number of pixel rows per frame H, and the number of pixels per row W of the pulse video encoded data; (2) Decode the pulse video encoded data according to the preset encoding method to obtain tag judgment data; (3) Allocate storage space, including: allocating two third buffer O[k][i] and fourth buffer N[i][k] with a capacity of W*H*N, to cache the original pulse video data and pixel bit string data, i =0,1,2,...,W*H-1, k=0,…,N-1; (3) Starting from k=1 and continuing to k=N, process the tag judgment data frame by frame, including: (3.1) When k<=3, the tag judgment data is the original pulse video data. The tag judgment data is read frame by frame and stored in O[k-1][i] in time sorting order, i =0,1,2,...,W*H-1; (3.2) When k=3, arrange the pixel data of the same pixel position in the first 3 frames of the label judgment data into a string according to the time order to obtain the pixel bit string of that pixel position, and store it in N[i][j], i =0,1,2,...,W*H-1, j =0,1,2; divide each pixel bit string into several micro strings, the micro string is a 1 in the pixel bit string and several 0s before it, until the preceding 1 or no preceding pixel; (3.3) When k>3, for each pixel position in the frame, determine the number n of microstrings in the pixel bit string corresponding to that pixel position up to the previous frame. i , where i represents the pixel position; If n i If the value is >=3, then feature extraction is performed on the pixel location, and the trained decision tree prediction model is called to predict the pixel based on the extracted features to obtain the predicted value Pre. If the label judgment data corresponding to the pixel location is 0, then O[k-1][i]=Pre; otherwise, O[k-1][i]=-Pre. If n i <3, the label judgment data is the original pulse video data, and the label judgment data of the pixel position in the current frame is directly stored in O[k-1][i]; Add the original pulse video data O[k-1][i] to the corresponding position in the pixel bit string; (3.4) k = k + 1, repeat steps (3.3) - (3.4) until k = N.

8. The lossless decoding method for pulse video data based on a decision tree prediction model according to claim 7, characterized in that, The feature extraction in step (3.3) specifically involves: Includes 5 features: feature First, the length of the nearest complete microstring to this pixel location, denoted as feat1; Feature 2 is the number of previously accumulated 0s in the microstring containing the pixel position, denoted as feat2; Feature 3: feat2 minus feat1; Feature 4 is the difference in length between the two microstrings preceding the pixel position; Feature 5 is the length difference between the second and third microstrings preceding the pixel position.

9. The lossless coding method for pulse video data based on a decision tree prediction model according to claim 7, characterized in that: The preset encoding method involves sequentially performing 0th-order exponential Columbus decoding and LWZ data compression decoding.

10. The lossless coding method for pulse video data based on a decision tree prediction model according to claim 9, characterized in that, The LWZ data compression encoding is specifically as follows: (2.1) Initialize the code table: Encode 257 zeros as 1, "1" as 2, "01" as 3, and so on. 256 zeros plus 1 1 are encoded as 258. When a new combination is encountered, the code value is accumulated from 259 downwards. (2.2) Obtain the length len of each frame from the decoded data of the 0th order exponent Columbus. i Frame by frame, LWZ decoding is performed by referring to the code table. If the last bit of the current frame data is 0, then (W*H - len) is added. i Add zeros to the current frame so that the data length of the current frame is W*H, where i represents the current frame number.

Images