METHOD FOR COMPRESSING IMAGE DATA

DE502019012477D1Active Publication Date: 2024-12-05ARNOLD & RICHTER CINE TECHNIK GMBH & CO BETRIEBS KG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
DE502019012477
Authority / Receiving Office
DE · DE
Patent Type
Patents
Current Assignee / Owner
Priority Date
2018-09-12
Filing Date
2019-09-12
Publication Date
2024-12-05
Estimated Expiration
2039-09-12

AI Technical Summary

Technical Problem

Current image data compression methods, especially lossy compression techniques, result in artifacts and require significant computational effort, while lossless methods struggle with efficiency and parallelization, particularly in high-resolution image processing for applications like digital cameras.

Method used

A method that converts pixel values into difference values using estimation rules, encoding these difference values with binary words of varying lengths, allowing for parallel encoding and decoding, thereby achieving high compression with minimal artifacts and efficient processing.

Benefits of technology

This approach enables fast, high-compression image data processing with reduced storage needs and real-time transmission capabilities, maintaining image quality by avoiding bit discarding and leveraging parallel processing for rapid encoding and decoding.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader
Need to check novelty before this filing date? Find Prior Art

Description

[0001] The invention relates to a method for compressing image data comprising a plurality of pixel values ​​assigned to a respective pixel.

[0002] Such image data is typically generated in an electronic camera, particularly a digital motion picture camera, by means of an image sensor. The image sensor usually comprises a multitude of light-sensitive sensor elements arranged in rows and columns, each corresponding to a pixel (picture element). Depending on the incident light, the sensor elements generate electrical signals corresponding to a specific brightness value. The sensor elements can be equipped with a pattern of color filters, such that each brightness value is assigned to a color channel, in particular red, green, and blue. The so-called Bayer pattern is known for this purpose, in which one red filter element, two green filter elements, and one blue filter element are arranged in a regular pattern.However, monochrome image sensors as well as image sensors with differently designed pixels are also suitable, especially sensors with multiple color channels, e.g. RGB-W or RGB-IR sensors.

[0003] With increasing spatial resolution, i.e., with an increasing number of pixels per image, the volume of image data in modern cameras has increased considerably. Therefore, compression methods are used to compress the image data generated in the camera and write it in compressed form to the camera's internal memory or output it to external storage. The compressed image data can later be decompressed. Lossless and lossy compression methods are used for this purpose. Like any application of data compression, image data compression is based on converting the original data, i.e., the pixel values, either into a fully reconstructible form that requires less storage space (lossless) or into a partially reconstructible form that also requires little storage space but does not contain all the data (lossy).Lossy compression involves removing data, but the loss should be barely perceptible. Image data compression can include, in particular, a prediction method and entropy coding. Additionally, preprocessing steps may be included, which can be lossy and thus result in lossy compression overall (even if the aforementioned combination of prediction and entropy coding is lossless). A requantization method can also be used. Methods for compressing image data using prediction and entropy coding are known, for example, from US 2017 / 070750 A1 and US 6,744,929 B1.

[0004] Furthermore, it is known to transform image data into the frequency domain (e.g., using discrete cosine transforms according to JPEG or wavelets according to JPEG 2000), where the frequency domain representation is achieved through coefficients that each describe the amplitude of specific frequencies. To compress the data, these coefficients are quantized as a function of frequency, whereby the coefficients corresponding to high frequencies are either inaccurately transferred or discarded. This removes fine and high-frequency structures in the image. Images compressed in this way therefore often have a very smooth, washed-out, and slightly blurry appearance. The image errors that arise during compression are difficult to predict and are not limited to individual pixels, but appear as additional structures (so-called artifacts).Furthermore, the transformation into the frequency domain is associated with high computational effort, so that fast implementations, such as those required for moving image cameras, involve a correspondingly high technical effort.

[0005] Compression methods are also known in which pixel values ​​are encoded by a variable number of bits. This approach is based on the understanding that some values ​​occur more frequently than others in image data. The frequent values ​​can be advantageously encoded by fewer bits. More bits can be allocated for the rarer values, but due to their low frequency, these only increase the total data volume negligibly. This approach is also known as entropy coding. One problem with this type of coding is its relatively poor parallelizability. This means that, due to the variable number of bits, the pixel values ​​can only be encoded (and decoded) sequentially, and therefore at a comparatively slow speed.

[0006] A coding method with variable codeword length is known from the internet publications Anonymous: "Universal Coding (part III) | Harder, Better, Faster, Stronger", (20140218), URL: https: / / hbfs.wordpress.com / 2014 / 02 / 18 / universal-coding-part-iii / , (20200123) and Anonymous: "Variablelength quantity - Wikipedia", (20180508), URL: https: / / en.wikipedia.org / w / index.php?title=Variable-length_quantity&oldid=840201447, (20200123).

[0007] The object of the invention is to provide a faster compression method that achieves good quality decompressed images at a high compression factor and avoids the formation of artifacts. Preferably, the compression method should also be lossless.

[0008] This problem is solved by a method having the features of claim 1.

[0009] In the method according to claim 1, the pixel values ​​for at least some pixels are compressed according to the following steps: Determining an estimated pixel value of the pixel in question as a function of the pixel value of at least one predetermined other pixel using at least one estimation rule; forming a difference value of the pixel in question corresponding to a predetermined relation between the pixel value and the estimated pixel value of the pixel in question, wherein the difference value is represented by a first binary word having a first length, the first length being an effective length of the first binary word; encoding the first binary word by at least one second binary word, wherein the at least one second binary word has a second length being a predetermined length of the respective second binary word. wherein the first length is compared with the second length to determine if the first length is smaller than the second length, wherein the first binary word is fully encoded by a single second binary word if the first length is smaller than the second length, and wherein otherwise the first binary word is fully encoded by multiple second binary words.

[0010] In summary, the pixel values ​​are first converted into difference values, which are then each encoded by at least one binary word of a predetermined length. Each difference value is encoded either by a single binary word or by multiple binary words, each of which has a predetermined length, which can be a fixed length. This predetermined length allows the encoding process for multiple difference values ​​to be performed in parallel, significantly accelerating the overall encoding process, especially when using graphics processors that are particularly well-suited for fast, parallel data processing.

[0011] The estimated pixel values ​​are preferably determined using a prediction method, in which a given pixel value is estimated or predicted, for example, by a weighted linear combination of predetermined other pixel values ​​(estimation rule). The difference values ​​thus correspond to an estimation or prediction error, which is generally significantly smaller than the underlying pixel value. Consequently, the difference values ​​cover a smaller range of values ​​than the pixel values ​​themselves, or are distributed more narrowly closer to zero. This allows the difference values ​​to be advantageously encoded in binary using a smaller number of bits, thus enabling the overall image data to be encoded with fewer bits and therefore compressed.

[0012] The difference value of the pixel in question is generally in binary form; that is, the difference value is represented by a number of bits of a first binary word, which can take on either the value zero or the value one. The difference value preferably corresponds to a decimal number represented by the first binary word in the so-called binary number space. Within the procedure, the difference value can initially be in decimal form and then transformed into a binary number to obtain the first binary word. The difference value can also be quantized and further processed. However, it is also possible for the difference value to be determined directly as the first binary word, for example, if the pixel value in question and the estimated pixel value are each represented and processed by binary words.Regardless of the specific method used to determine the first binary word, it can be considered an intermediate encoding of the underlying pixel value. The first binary word is then further encoded by at least one second binary word, as described in more detail below.

[0013] The first binary word has an effective initial length, which preferably depends on the magnitude of the difference value. In general, the effective length can correspond to the number of bits provided in the binary number system for representing a decimal number. If more bits are provided, these are preferably set to zero (for example, when executing the method on a computer that only processes binary words of a fixed length). For instance, the first binary word, in an ascending order of bits after the last bit, which is set to one ("most significant bit"), can contain several bits, each set to zero. The effective length thus corresponds to the number of bits between and including the least significant bit and the most significant bit; that is, the least significant bit and the most significant bit are counted.

[0014] The first length of the first binary word preferably increases with the size of the difference value represented by the first binary word. Thus, the first length depends on the difference value, so that large and therefore rare difference values ​​are represented by more bits than small or frequent difference values. The first binary word is then encoded by one or more second binary words, each with a predetermined second length, thereby enabling advantageous parallel encoding of multiple first binary words. Preferably, each of the multiple second binary words has the second length, i.e., the second length is identical for all second binary words. This maximizes the advantage of parallel encoding. However, as will be explained later, the multiple second binary words can also have different predetermined second lengths.

[0015] Depending on the length of the first binary word, each first binary word is encoded either by a single second binary word or by multiple second binary words. If the first length is less than the predetermined second length, the second binary word has enough bits available to encode the first binary word completely and therefore without loss. If the first length is greater than or equal to the second length, the first binary word is encoded by multiple second binary words; that is, the information of the first binary word is distributed across several second binary words to encode the first binary word completely and therefore without loss. In this way, the first binary word is adaptively encoded depending on the length of the first binary word. The advantage of a variable first length, which significantly compresses the image data, is combined with the advantage of a predetermined and, in particular, fixed word length for the second binary words.This creates a compression method that can be performed particularly quickly, even at high compression levels. Furthermore, no bits need to be discarded during encoding, ensuring high quality in the decompressed image data and preventing artifacts. The high compression level allows for reduced storage space usage and exceptionally fast transfer of the compressed data to a remote server, such as a virtual cloud. Additionally, the data can be quickly encoded after pixel values ​​are recorded, enabling real-time transmission to a server. This also reduces the storage resources required by devices like video cameras.Overall, the invention thus makes a significant contribution to the globally networked society, in which current image data should be available immediately and everywhere.

[0016] The second length is the same for each pixel group, where each pixel group represents a local area within the image data. Having a second length that is only the same for each pixel group has the advantage that it can be varied depending on the image content to achieve an improved compression ratio. For example, with a relatively large number of long first lengths, the improved compression ratio can be achieved by increasing the second length, thus requiring fewer encoding information bits overall. The information about the second length can be integrated into the compressed image data for decompression and, for example, transmitted to a decoder in a bitstream representing the compressed image data.A similar approach can be provided if the first binary word is fully encoded by several second binary words and the second lengths for the several second binary words are different, whereby the different second lengths can then be uniformly predetermined for a respective pixel group.

[0017] The second length is chosen proportionally to the average first length of the first binary words of the pixel group in question.

[0018] The second binary words used to encode a given first binary word can be concatenated in a bitstream. The concatenation order can correspond to the order of the first binary words considered for a given encoding. However, it is possible to specify a reverse order for all or even just a subset of the second binary words. In particular, if encoding is performed using multiple second binary words, these multiple second binary words can be written into a bitstream in reverse order. This has the advantage that, during decoding, it is not necessary to read all multiple second binary words before beginning the formation of the first binary word. For example, the order of the second binary words during encoding can correspond to the order of the bit weights of the first binary word (i.e.,(starting with the least significant bit and ending with the most significant bit of the first binary word). The second binary words can then be written into the bitstream in reverse order. Individual second binary words, however, can be written in the regular order.

[0019] Within the context of the disclosure, the term "ascending order" is generally to be understood as a predetermined or defined sequence. This sequence can be defined by a "reading direction" of bits, e.g., from left to right, or by an order of the weight of bits in a given binary word (e.g., from right to left).

[0020] In general, it is not necessary to compress all pixel values ​​of a given image data set, which may, for example, correspond to a single image of an image sequence, according to the method according to the invention. Instead, the method according to the invention can also be performed only on some of the pixel values. For example, the compression can be preceded by an analysis of the image data, whereby, as a result of the analysis, a first compression method (in particular the method according to the invention) is used for some (variable or predetermined) image areas, and a different second compression method is used for other image areas, particularly to achieve a higher compression rate and / or lower storage requirements at the expense of higher computational effort and / or poorer parallelizability.

[0021] The invention is generally not limited to a specific order of compression of the pixel values ​​or of the arrangement of the assigned pixels.

[0022] Within the scope of the invention, it is also not excluded that preprocessing steps (e.g. quantization, linearization, gamma curve) are carried out after the generation of the original image data (for example by means of an image sensor) and before the compression of the pixel values.

[0023] Advantageous embodiments are disclosed in the dependent claims, the description and the figures.

[0024] According to a preferred embodiment, the at least one second binary word comprises at least one encoding information bit, wherein the encoding information bit is set to a first value if the first binary word is encoded by a single second binary word, and wherein otherwise the encoding information bit of at least one of the several second binary words is set to a second value. The encoding information bit thus indicates how the first binary word was encoded. This information can be used, in particular, during subsequent decoding to completely reconstruct the first binary words. Preferably, only one encoding information bit is provided in every second binary word, which makes the encoding (and also the decoding) process particularly simple and therefore reliable. However, other methods are also possible.In particular, multiple encoding information bits can be used to provide more complex information about the encoding process for decoding. Furthermore, it is conceivable that, especially when the first binary word is encoded by several second binary words, not all second binary words contain an encoding information bit. For example, encoding information can be restricted to one or more specific second binary words.

[0025] Knowing the length of the second binary word(s) allows for the direct determination of a starting point during subsequent decoding. However, it is also possible to determine the beginning of a pixel, or the corresponding second binary word(s), by evaluating the coding information bits. Furthermore, it is possible that coding information bits are not part of the second binary words themselves. For example, coding information bits can be stored separately from the second binary words, such as in a header file, and then considered during subsequent decoding using a mapping table.

[0026] Within the scope of this application, the aforementioned first value of the coding information bit is preferably always zero, and the aforementioned second value of the coding information bit is preferably always one. However, a reversal of this assignment is also possible. Furthermore, it is understood that setting a bit means ensuring that the bit in question has the desired value. For this purpose, it can be explicitly or implicitly checked whether the bit already has the desired value before any potential setting of the bit. If so, the bit does not need to be set again, although this is possible for reasons of reliability. It is also understood that in the binary system, a bit can only assume one of two values. Thus, checking whether the bit is set to, for example, the first value can also be done by checking whether the bit is set to the second value.

[0027] Preferably, the first binary word is encoded by the at least one second binary word by adding the encoding information bit to the first binary word or a subword of the first binary word to form the at least one second binary word, wherein the encoding information bit has a predetermined position in the second binary word. The predetermined position can, in particular, be an outer or edge position in the second binary word. Preferably, the position of the encoding information bit corresponds to the position of the least significant bit in the second binary word. For this purpose, the encoding information bit can be appended to the first binary word or subword, for example, on the side of the least significant bit of the word in question. This can, in particular, speed up the decoding process.

[0028] If the sum of the first length and the number of at least one encoding information bits is less than the second length, the second binary word can be formed by concatenating the first binary word with a number of zero-valued bits, such that the at least one second binary word has the second length. Preferably, the zero-valued bits are appended to the first binary word next to its most significant bit (one), so that the bit pattern of the first binary word remains unchanged. This makes the first binary word particularly easy to decode later.

[0029] The number of zero-valued bits depends on the number of encoding information bits (preferably one) and the length of the first bit. The number of zero-valued bits can be determined by subtracting the length of the second bit from the number of encoding information bits and the length of the first bit.

[0030] It is understood that the procedural steps described here can generally be carried out in different ways. For example, the aforementioned or any other appending of zero-valued bits can be achieved by overwriting a portion of the bits in a second binary word, in which all bits are set to zero, according to the bits of the first binary word or subword. Similarly, the encoding information bit can be added by setting a designated bit in the second binary word to the value of the encoding information bit, thereby making it the encoding information bit. In other words, at least some of the procedural steps described here should also be understood in terms of the concept or result implemented by each respective step.

[0031] In one embodiment, the first binary word is encoded by the single second binary word by concatenating the first binary word and the encoding information bit to form the single second binary word. The encoding is thus particularly simple, and the bit pattern or bit sequence of the first binary word remains unchanged. Preferably, an encoding information bit is appended to the first binary word on the side of the least significant bit.

[0032] In a further embodiment, the first binary word is encoded by the multiple second binary words by decomposing or splitting the first binary word into several subwords. Each of these subwords has a third length, which is a predetermined, in particular fixed, length of the respective subword and shorter than the second length. Furthermore, each of the multiple subwords is assigned one of the multiple second binary words, and each subword is concatenated with the encoding information bit of the assigned second binary word to form the assigned second binary word. Preferably, exactly one second binary word is provided for each subword in order to make the encoding and subsequent decoding efficient and particularly robust.

[0033] The decomposition of the first binary word into its multiple subwords preferably occurs in ascending order of the bits of the first binary word, with each subword being formed by a number of bits from the first binary word equal to the third length. If the third length is not an integer multiple of the first length, the last of the multiple subwords in ascending order can be formed by concatenating a number of zero-valued bits with a remainder of bits from the first binary word, such that the sum of the number of zero-valued bits and the remainder of bits from the first binary word equals the third length. The subwords thus correspond to contiguous bit sequences or blocks from the first binary word that are intended to have the third length. Provided the first length is divisible by the third length without a remainder, the subwords can be formed directly and completely by blocks of the first binary word.In the event of a remainder, the remaining bits of the first binary word are supplemented with zero-valued bits to form the last subword. In this way, the encoding can be performed completely and without loss even if the first binary word cannot be exactly split into blocks of the third length.

[0034] Generally, the ascending order of bits in a binary word, especially in the first binary word, is the order of bits starting with the least significant bit and progressing to the most significant bit. However, other definitions are conceivable in principle.

[0035] According to one embodiment, the coding information bit, which is concatenated with the last of the multiple subwords in ascending order, is set to the first value (in particular, one), while the respective information bit of the remaining subwords is set to the second value (in particular, zero). In this case, the coding information bit indicates in every second binary word whether the first binary word has already been completely encoded by the preceding second binary words or not. By evaluating successive coding information bits (of immediately consecutive second binary words), it is therefore possible to immediately determine which second binary words underlie a common first binary word and are accordingly to be used for decoding this first binary word. The image data can thus be compressed and decompressed particularly efficiently and reliably.

[0036] Preferably, the second and third lengths differ by exactly the number of at least one encoding information bits in order to maximize the compression factor. However, it is also possible to include additional bits in the second binary word, for example for error correction, which correspondingly reduces the available third length.

[0037] According to one embodiment, several pixel groups can be formed depending on the similarity of the pixel difference values, such that the pixel difference values ​​of a given pixel group are more similar to each other than to the pixel difference values ​​of another pixel group. For each pixel group, a second length can then be chosen that is optimal for encoding the difference values ​​or first binary words in terms of a high compression ratio.

[0038] As previously explained, the second length is proportional to the size of the difference values ​​of a given pixel group. For simplicity, the second length can be chosen to be proportional to the first length of the first binary words representing the difference values ​​of that pixel group. For example, the second length can be chosen based on the average first length of the first binary words belonging to that pixel group. Thus, it is conceivable that for a given pixel group, the second length could be, for instance, equal to the average first length minus the number of encoding information bits.

[0039] The predetermined length of each second binary word can be constant for multiple second binary words if the respective first binary word is fully encoded by several second binary words. This can increase the advantage of parallel processing.

[0040] However, it is also possible that—if the first binary word is fully encoded by several second binary words—the second lengths of these second binary words will differ. For example, the second length can decrease with an increasing number of second binary words used to encode the first binary word. To determine the respective second length, an indirect local evaluation of the image content can be performed. This involves encoding the first binary words of an image area with different values ​​for the second length of the second binary words and evaluating the resulting coding efficiency. The second length(s) with the highest coding efficiency are then selected, and this evaluation result is stored in a control table (for later decompression).It is also conceivable that statistics of pixel and / or difference values ​​are taken into account in such a way that a fixed length is used in most cases and a different length is used only in rare cases.

[0041] In a further embodiment, the step of encoding the first binary word with at least one second binary word corresponds to a first encoding method, wherein for at least some further pixels that are not encoded according to the first encoding method, the first binary word is encoded according to a second encoding method. In other words, as an alternative to the first encoding method, a different encoding method can be provided for certain image areas or pixel groups; that is, a selection is made between at least two different encoding methods. The second encoding method can generally be designed arbitrarily, with a Golomb-Rice encoding method being particularly possible. The selection of a respective encoding method can, for example, be based on a local evaluation of the image content.

[0042] With regard to at least one predetermined other pixel whose pixel value is used to determine the estimated pixel value, it is possible to pursue one or more advantageous procedures, which are described below.

[0043] In one embodiment, the image data comprises pixel values ​​of several different colors, wherein the at least one predetermined other pixel is not located directly adjacent to the pixel in question, and wherein the pixel values ​​of the pixel in question and the at least one predetermined other pixel are of the same color. In other words, only pixel values ​​of the same color are used to estimate a pixel value. This can improve the accuracy of the estimate. Often, the color filters of the same color in an image sensor are not directly adjacent to each other. This is particularly true for the Bayer pattern mentioned earlier. It is understood that multiple pixels can be used to determine the estimated pixel value in order to increase the accuracy of the estimated pixel value.

[0044] In a further embodiment with pixel values ​​of several different colors, the estimated pixel value of the pixel in question can be determined based on the respective pixel values ​​of several predetermined other pixels. Some of these pixels are located directly adjacent to the pixel in question, and their pixel values ​​are a different color than the pixel value of the pixel in question. However, another portion of the predetermined pixels is not located directly adjacent to the pixel in question, and their pixel values ​​are the same color as the pixel value of the pixel in question. In other words, pixel values ​​of only one or more other colors are also used to determine the estimated pixel value.This allows correlations between the pixel values ​​of the same as well as different colors to be used to further increase the estimation accuracy.

[0045] In general, the pixels can be arranged in a multi-row configuration, with the pixel values ​​being compressed row by row. The estimated pixel value of the pixel in question is determined based on the respective pixel values ​​of several predetermined other pixels. These other pixels can be located in the same row as the pixel in question and / or in an already compressed row. Alternatively, the other pixels can be located exclusively in the same row as the pixel in question, which can further accelerate the process.

[0046] The image data can, in particular, represent a single frame from a temporal sequence of several frames; that is, the single frame can, for example, be taken from a video or moving image sequence. The pixel values ​​belonging to the single frame do not necessarily have to have been captured at the same time. The pixel values ​​can also have been captured sequentially and combined to form a single frame.

[0047] The method can generally be configured to compress image data losslessly. "Lossless" in this context means that the first binary words are fully represented by the second binary words without any loss of information and can be decoded without loss. Thus, the first binary words can also represent quantized difference values, with the information loss due to quantization not considered significant. Each difference value can therefore be quantized according to a difference-value quantization rule, with the first binary word representing the quantized difference value. Apart from rounding effects, completely lossless compression is also possible.

[0048] As mentioned above, the process can be performed particularly quickly by encoding the difference values ​​of several pixels in question, each represented by a first binary word, in parallel. This can be done, for example, by graphics processing units (GPUs), which are particularly well-suited for the parallel and block-wise processing of image data (parallel "threads"). Such GPUs are widely available on modern graphics cards and are becoming increasingly common.

[0049] The invention also relates to a method for compressing and decompressing image data (decompression method), which are compressed according to one of the embodiments of the compression method described above and which comprise a bitstream with a plurality of bits, wherein the bitstream represents a plurality of pixel values ​​that are assigned to a respective pixel, wherein the decompression method comprises the following steps: Decomposing the bitstream into a plurality of second binary words, wherein each of the second binary words has a second length which is a predetermined, in particular fixed, length of the respective second binary word, wherein each of the second binary words has at least one coding information bit at a predetermined position in the second binary word;Decoding at least one first binary word by at least one of the second binary words, wherein the first binary word represents a difference value corresponding to a predetermined relation between a pixel value of a pixel in question and an estimated pixel value of the pixel in question, determining whether the coding information bit of the at least one of the second binary words is set to a first value, wherein the first binary word is completely decoded by a single one of the second binary words if the coding information bit of the single one of the second binary words is set to the first value, and wherein otherwise the first binary word is completely decoded by several of the second binary words; determining a simulated estimated pixel value of the pixel in question as a function of at least one previously decompressed pixel value corresponding to the at least one predetermined other pixel, using an estimation rule;and determining a decompressed pixel value of the pixel in question as a function of the first decoded binary word and the simulated estimated pixel value of the pixel in question.

[0050] In summary, the decompression method is the reverse of the compression method described previously. Thus, the difference values, in the form of first binary words, are decoded either by a single second binary word or by multiple second binary words in the bitstream. For this purpose, the coding information bit contained in each second binary word is evaluated, indicating whether a single second binary word or multiple second binary words are required for decoding to achieve a complete and lossless decoding.

[0051] Preferably, each second binary word contains an encoding information bit, so that a given second binary word can always be uniquely used for the correct decoding of a first binary word. However, other approaches are also possible, as explained above in the context of compression.

[0052] The decoding of the first binary word by the at least one second binary word is to be understood as meaning that the first binary word is determined based on the at least one second binary word. This can be interpreted as decoding the first binary word from the at least one second binary word. Conversely, the encoding of the first binary word by the at least one second binary word is to be understood as meaning that the at least one second binary word is determined based on the first binary word.

[0053] Determining the simulated estimated pixel value of the pixel in question is entirely analogous to determining the estimated pixel value during compression, whereby certain information, such as weighting factors for predetermined other pixels, can be provided, for example, as part of the estimation rule during decompression. Such information provision is known from the field of linear prediction.

[0054] In one embodiment of the decompression method, the first binary word is decoded by the at least one second binary word by removing the encoding information bit from the at least one second binary word to form the first binary word or a subword of the first binary word. The removal of a respective encoding information bit is preferably applied in both decoding cases, i.e., both when decoding by a single second binary word and when decoding by multiple second binary words.

[0055] It is possible that the encoding information bit in each second binary word always occupies the position of the least significant bit. After the encoding information bit has been removed, bits remain. These remaining bits can directly form the first binary word to be decoded or a subword thereof. Preferably, the remaining bits form the first binary word if the removed encoding information bit has the first value and the encoding information bit of the previous second binary word also has the first value. Otherwise, the bits form a subword. In the case of decoding by a single second binary word, the last bit set to one in the ascending order of the remaining bits can be the last, i.e., the most significant, bit. However, if there are still zero bits in the remaining bits after the last ones bit, these zero bits can be automatically truncated or...These null bits are removed so that the first binary word does not include them. In this way, the first binary word can be decoded with exactly the same length as the original length used in the previous encoding.

[0056] It is understood that during decompression, several initial binary words can be decoded in parallel, thus significantly accelerating the decompression of the entire image data. Graphics processors, for example, can be used for this purpose, just as in compression.

[0057] In another embodiment of the decompression method, the first binary word is decoded by the multiple second binary words by performing the following steps: Identifying a set of second binary words corresponding to the multiple second binary words used to decode the first binary word, wherein the second binary words of the set are identified by evaluating, starting with a relevant second binary word of the bitstream whose encoding information bit is not set to the first value and which forms a first element of the set, the respective encoding information bits of the subsequent second binary words in ascending order of the bitstream to determine whether the respective encoding information bits are set to the first value, wherein the subsequent second binary words form further elements of the set if the respective encoding information bits are not set to the first value, wherein the subsequent second binary word whose encoding information bit is set to the first value forms the last element of the set, forming multiple subwords.by removing at least the encoding information bit from each of the second binary words in the set, and forming the first binary word by concatenating the several subwords together.

[0058] It is understood that identifying the set of multiple second binary words, forming the subwords, and forming the first binary word do not necessarily have to be performed sequentially for all involved words. Thus, it is possible for the second binary words to be processed at least partially in parallel in order to decode each first binary word. Furthermore, second binary words can be evaluated sequentially in their bitstream order with respect to their respective coding information bit to determine whether or not further subsequent second binary words need to be considered. If the evaluation of the coding information bit reveals that at least one further subsequent second binary word needs to be considered, a subword can be formed and stored by removing the relevant coding information bit.Subsequent subwords can be stored in the same way, whereby these subwords can already be concatenated to successively form the first binary word by a series of subwords. In other words, the steps for decoding successive second binary words can also be alternately nested within each other.

[0059] Preferably, the subwords are concatenated in an order that is the reverse of the ascending order of the bitstream in which the subwords appear. For example, the first element of the set (a second binary word with an encoding information bit not set to the first value) can contain the least significant bits for the first binary word to be decoded. Conversely, the last element of the set (a second binary word with an encoding information bit set to the first value) can contain the most significant bits for the first binary word to be decoded. Accordingly, the set can have an "identification order" that corresponds to the ascending order of the second binary words in the bitstream in which the second binary words were identified. In contrast, the formed subwords can be concatenated or strung together in reverse order.This has the advantage that, on the one hand, a given first binary word can be sequentially encoded by the multiple second binary words; that is, it does not need to be known in advance how many second binary words are required. On the other hand, the reverse order of the second binary words in the bitstream means that it can be determined immediately from a "first" second binary word (the first element of the set) that multiple second binary words are needed for decoding. Here, too, it does not need to be immediately clear how many second binary words are required. Rather, the set of second binary words can be sequentially identified by evaluating the respective encoding information bits.

[0060] As mentioned above, in addition to the coding information bit from the last of the formed subwords, zero-valued bits present after the last bit set to one ("most significant bit") can be automatically removed. However, this is not strictly necessary.

[0061] It is understood that features disclosed in connection with the compression process may be implemented in the decompression process in a corresponding or inverse manner. Conversely, features disclosed in connection with the decompression process may also be implemented in the compression process in a corresponding or inverse manner. Furthermore, the compression and decompression processes may also be carried out sequentially, i.e., the embodiments may also be combined.

[0062] The invention further relates to a data carrier (storage medium) with software stored on the data carrier for carrying out the method according to one of the embodiments of the compression method and / or the decompression method described herein. The software may also be firmware, i.e., embedded software.

[0063] The invention further relates to a device comprising a data carrier, in particular the aforementioned data carrier, and a computing unit, wherein the computing unit is adapted to execute the software stored on the data carrier. The computing unit may include at least one graphics processor and / or be connected to at least one graphics processor, in particular to execute the software or to operate the computing unit according to the software. Furthermore, the software may generally be implemented by a field-programmable gate array (FPGA). It is understood, however, that the computing unit need not necessarily include these means and that the method can also be executed efficiently in other ways.

[0064] The device can generally be configured to record a temporal sequence of individual images. The device can, in particular, be a film camera. The processing unit of the device is preferably configured to compress at least one of the individual images according to the compression method described above. With regard to decompression of the image data, the processing unit can also be configured to decompress a bitstream representing at least one of the individual images according to the decompression method described above. This can be particularly advantageous in a film camera, so that a given individual image can, for example, be displayed directly on a display device of the device or transmitted to an external display device.The image data does not need to be stored in uncompressed form; it can be compressed immediately after recording. Preferably, the image data is transmitted in compressed form, thus requiring little bandwidth for transmission (fast transmission) and little storage space. For example, the compressed image data can easily be transferred to a remote server that hosts a virtual cloud for image data. The methods described here even allow for real-time compression and decompression, enabling, for instance, individual images to be compressed and transmitted remotely immediately after capture.

[0065] The methods can generally also be implemented in other devices, e.g. in a digital microscope, which can be used particularly in the medical field, for example to perform operations.

[0066] In general, the invention also relates to software on a data carrier, wherein the software is configured to carry out at least one of the methods described herein.

[0067] The invention is explained below by way of example only, with reference to the drawings, wherein Fig. 1 shows a schematic diagram illustrating the general encoding of a first binary word by at least one second binary word; Fig. 2 shows an example of encoding a first binary word by a single second binary word; Fig. 3 shows an example of encoding a first binary word by several second binary words; Fig. 4 shows an example of decoding several first binary words by several second binary words of a bitstream.

[0068] In the following, identical or corresponding elements will be marked by the same reference symbols unless a special distinction between specific instances of elements is necessary.

[0069] In Fig. 1 Figure 1 illustrates a method for compressing image data in which a first binary word 12 is encoded by at least one second binary word 14, 14'. The first binary word 12 represents a difference value of a particular pixel in the image data, where the difference value corresponds to a predetermined relation between a pixel value and an estimated pixel value of the particular pixel (not shown). For example, the difference value can represent the difference between the (actual) pixel value and the estimated pixel value. The estimated pixel value of the particular pixel can be determined as a function of the pixel value of at least one other predetermined pixel in the image data using at least one estimation rule (not shown). For example, the estimated pixel value can be formed as a weighted combination of the pixel values ​​of other pixels, where the weighting coefficients are, for example,This can be determined by minimizing the mean squared error between the pixel value and the estimated pixel value of the pixel in question. It is understood that the image data comprises a multitude of pixel values, each assigned to a specific pixel. Thus, at least some of the pixel values ​​can be converted into an associated difference value, which either exists directly in the form of a first binary word (especially when the procedure is executed on a binary computer) or is converted into a first binary word from another number system. Therefore, several first binary words must be encoded to compress the image data. For the sake of clarity, only one such first binary word, 12, will be considered in the following discussion.

[0070] The first binary word 12 has a first length L1, which is the first effective length of the first binary word 10. If the first binary word 12 is, for example, (001011), then the first length L1 is 4, since, starting with the least significant bit (the rightmost bit in the bit sequence 001011), the last ones bit is in the fourth position and thus, in particular, represents the most significant bit in the binary system. If the first binary word 12 is (101100), then the first length L1 is 6, since the last ones bit is in the sixth position (leftmost in the bit sequence 101100). The first length L1 is therefore generally variable, depending on the position of the last ones bit in a binary word. This is indicated by the dashed arrow 16.

[0071] Depending on the length L1, the first binary word is encoded either by a single second binary word 14 or by multiple second binary words 14'. Encoding is done by a single second binary word 14 if the first length L1 is smaller than the second length, i.e., L1 ≤ 1. <L2 (linker Zweig in Fig. 1 The encoding, however, is performed by the multiple second binary words 14' if the first length L1 is greater than or equal to the second length, i.e., L1 > L2 or L1 = L2 (right branch in Fig. 1 ).

[0072] The encoding by the single second binary word 14 (case L1) <L2) erfolgt vorzugsweise durch Verkettung des erstes Binärworts 12 mit einem einzelnen Kodierinformationsbit 18, das auf den Wert Null gesetzt wird und auf diese Weise anzeigt, dass das betreffende erste Binärwort 12 vollständig durch ein einzelnes zweites Binärwort 14 kodiert worden ist. Das zweite Binärwort 14 besitzt definitionsgemäß die zweite Länge L2, sodass die Differenz zwischen der zweiten Länge L2 und der Anzahl der Kodierinformationsbits 18 allgemein eine dritte Länge L3 ist, welche kleiner als die zweite Länge L2 ist. Da in Fig. 1 Assuming, for example, a single coding information bit 18, then L3=L2-1 .

[0073] If the first length L1 is equal to the third length L3, the encoding of the first binary word 12 is simply a concatenation or sequence with the encoding information bit 18. However, if the first length L1 is less than the third length L3, the first binary word 12 is appended with a number of zeros, so that the second binary word 14, after concatenation with the encoding information bit 18, has a second length L2. Preferably, the zero bits are appended to the side of the most significant bit. It is also possible to define a second binary word 14 with zero bits of length L2 and to overwrite a portion of these zero bits according to the encoding information bit 18 and the L1 bits of the first binary word 12.

[0074] In the case of encoding by several second binary words 14' (case L1 >L2 or L1=L2), the first binary word 12 is first decomposed into several subwords 20 of length L3, each of which is concatenated with a coding information bit 18 to form the second binary words 14' (cf. Fig. 1 The coding information bits 18 are usually set to one to indicate that the first binary word 12 has not yet been fully encoded. An exception to this is the last second binary word 22 according to an ascending sequence of second binary words 14' (see...). Fig. 1 ), which differs from the other second binary words 14' in that the coding information bit 18 is set to zero. This indicates that the first binary word 12 has been fully encoded with the last second binary word 22 and the preceding second binary words 14'.

[0075] It goes without saying that the values ​​for the coding information bit 18 can also be inverted.

[0076] The coding procedure is further explained below using concrete numerical examples.

[0077] In Fig. 2 The calculation starts with a difference value of 24, which has the value 11 in the decimal system. The difference value 24 is represented by a first binary word 12', which is in Fig. 2 Although it comprises 12 bits, it has an effective length L1=4, since the last ones bit in the ascending order of the bits is the fourth bit (step 32 in Fig. 2 ). The ascending order is in Fig. 2 defined from right to left, i.e. from the least significant bit 26 to the most significant bit 28.

[0078] Those bits of the first binary word 12' that define the first length L1 now form a first binary word 12, which is used as the basis for the encoding (step 34 in Fig. 2 ). As from Fig. 2 As can be seen, the first binary word 12 and the first binary word 12' differ only by the zeros present beyond the last ones bit in the first binary word 12'.

[0079] The first binary word 12 is now encoded by a single second binary word 14, which has the second length L2=6 (step 36 in Fig. 2 ). Since L1 <L2 ist, wird das Kodierinformationsbit 18, welches mit dem ersten Binärwort 12 an der Seite des niederwertigsten Bits 26 verkettet wird, auf den Wert Null gesetzt. Da L2-1> Since L1 is the length of the first binary word 12, a zero bit 30 is appended to the side of the most significant bit 28, so that the second binary word 14 has a length of L2. In other words, the first binary word 12 is supplemented with a number of zero bits 30 to ensure that the second length L2 is independent of the first length L1.

[0080] In Fig. 3 Another encoding example is shown, which assumes a higher difference value 24', which has the value 868 in the decimal system and is represented by a first binary word 12', which comprises 12 bits but has an effective length L1=10, since the last ones bit in the ascending order of the bits, 28, is the tenth bit (step 38 in Fig. 3 A first binary word 12 with effective length L1 is obtained in step 40 by truncating the zero bits following the most significant bit 28. Since, as in Fig. 2 Since encoding is to be performed using second binary words 14' of length L2=6, the first binary word 12 is encoded using multiple second binary words 14' because L1>L2. For this purpose, the first binary word 12 is decomposed into subwords 20 (step 40 in Fig. 3 ), wherein in the ascending order of the bits of the first binary word 12, L3 bits are combined to form a subword 20 (steps 40 and 42 in Fig. 3 ). In Fig. 3 The binary word 12 can be completely decomposed into two subwords 20 of length L3. If the length L1 is not divisible by L3 without a remainder, the remaining bits are represented by another subword 20, which is supplemented with zero bits, so that the subword 20 in question also has a length of L3 (not shown).

[0081] The 20 sub-words are now each processed analogously as in the example of the Fig. 2 concatenated with a coding information bit 18, where the coding information bit 18 of the first of the two subwords 20 (right in Fig. 3 ) is set to one (first binary word 12' is not yet fully encoded) and where the encoding information bit 18 of the second of the two subwords 20 (left in Fig. 3 ) is set to zero (first binary word 12' is fully encoded), see step 44 in Fig. 3 In this way, several second binary words 14' are generated, by which the first binary word 12' is encoded. The second, or leftmost, of the two binary words 14' corresponds to the last second binary word 22 in Fig. 1 The two second binary words 14' can then be concatenated to form a bitstream 50 or a part of a bitstream, see step 46 in Fig. 3 It is also possible to reverse the order of the two second binary words 14'. If the bitstream 50 is read from left to right, for example, the first encoding information bit 18 in this sequence, which then has the value one, already indicates that further second binary words 14' must be considered for decoding.

[0082] In Fig. 4 Figure 5 is an example of decoding several first binary words 12' by several second binary words 14, 14', where the decoding is part of a procedure for decompressing image data comprising a bitstream 50' with a plurality of bits. The bitstream 50' is decomposed into several second binary words 14, 14', each with a length of L2=6 (step 52). The bitstream 50' is read in ascending order, with the bits starting with the leftmost bit of the bitstream 50' in Fig. 4 The data is read successively up to the rightmost bit. However, a reverse order is also possible. Then, the encoding information bit 18 of each of the second binary words 14, 14' is evaluated. For the second binary words 14, the encoding information bit is zero, meaning that each second binary word 14 serves to decode a single first binary word 12', for which only the encoding information bit 18 is removed (steps 54 and 56).

[0083] In Fig. 4 The coding information bit 18 of the second binary word 14' located in the middle of the bit stream 50' has the value one. This indicates that a first binary word 12 is to be decoded by this second binary word 14' and at least one further second binary word 14'. The coding information bit 18 of the subsequent second binary word 14' again has the value zero. This indicates that only this subsequent second binary word 14' needs to be considered to decode the first binary word 12'. In this way, a set of second binary words 14' is identified for decoding the first binary word 12'. Subwords 20 are then formed by removing the coding information bits 18 from the second binary words 14' (step 54). The part words are then concatenated, in reverse order of the part words 20. Since in the example of Fig. 4 Since only two subwords 20 are present, these are first swapped and then concatenated (step 56). The swapping results from the fact that the first second binary word 14' in the ascending order, whose encoding information bit 18 has the value one, contains lower-order bits of the first binary word to be decoded, whereas the following second binary word 14' contains higher-order bits.

[0084] The decoded first binary words 12' can then be converted into difference values ​​24,24' and transformed into decompressed pixel values ​​based on simulated estimated pixel values. The decompressed pixel values ​​can then be used on a display device to show a single image to a user. Bezugszeichenliste

[0085] 12, 12' first binary word 14, 14' second binary word 16 length 18 encoding information bit 20 subword 22 last second binary word 24, 24' difference value 26 least significant bit 28 most significant bit 30 zero bit 32, 34, 36 step 38, 40, 42 step 44, 46 step 50, 50' bit stream 52, 54, 56 step L1 first length L2 second length L3 third length

Claims

1. A method of compressing image data that comprise a plurality of pixel values which are associated with a respective pixel, wherein the pixel values are compressed for at least some pixels according to the following steps: - determining an estimated pixel value of the respective pixel in dependence on the pixel value of at least one predetermined other pixel while using at least one estimation rule; - forming a difference value (24, 24') of the respective pixel that corresponds to a predetermined relation between the pixel value and the estimated pixel value of the respective pixel, with the difference value (24, 24') being represented by a first binary word (12, 12') which has a first length (L1), with the first length (L1) being an effective length of the first binary word (12, 12'), characterized by: - encoding the first binary word (12, 12') by at least one second binary word (14, 14'), with the at least one second binary word (14, 14') having a second length (L2) that is a predetermined length of the respective second binary word (14, 14'), wherein the first length (L1) is compared with the second length (L2) to determine whether the first length (L1) is smaller than the second length (L2); wherein the first binary word (12, 12') is completely encoded by a single second binary word (14) when the first length (L1) is smaller than the second length (L2); and wherein the first binary word (12, 12') is otherwise completely encoded by a plurality of second binary words (14'), wherein the second length (L2) is the same for a respective pixel group, wherein the respective pixel group represents a local image region in the image data, and wherein the second length (L2) is selected as proportional to the average first length of the first binary words (L1) of the respective pixel group.

2. A method according to claim 1, wherein the at least one second binary word (14, 14') comprises at least one encoding information bit (18); wherein the encoding information bit (18) is set to a first value when the first binary word (12, 12') is encoded by a single second binary word (14); and wherein otherwise the encoding information bit (18) of at least one of the plurality of second binary words (14') is set to a second value.

3. A method according to claim 2, wherein the first binary word (12, 12') is encoded by the at least one second binary word (14, 14') in that the encoding information bit (18) is appended to the first binary word (12, 12') or to a partial word (20) of the first binary word (12, 12') to form the at least one second binary word (14, 14'), wherein the encoding information bit (18) has a predetermined position in the second binary word (14, 14'), wherein, preferably, if the sum of the first length (L1) and the number of the at least one encoding information bit (18) is smaller than the second length (L2), the first binary word (12, 12') is concatenated with a number of zero-value bits (30) so that the at least one second binary word (14, 14') has the second length (L2); and / or wherein the first binary word (12, 12') is encoded by the single second binary word (14) in that the first binary word (12, 12') and the encoding information bit (18) are concatenated with one another to form the single second binary word (14).

4. A method according to claim 2 or 3, wherein the first binary word (12, 12') is encoded by the plurality of second binary words (14') in that the first binary word (12, 12') is segmented into a plurality of partial words (20); wherein each of the plurality of partial words (20) has a third length (L3) that is a predetermined length of a respective partial word (20) and that is smaller than the second length (L2); wherein one of the plurality of second binary words (14') is associated with each of the plurality of partial words (20); and wherein each of the plurality of partial words (20) is concatenated with the encoding information bit (18) of the associated second binary word (14') to form the associated second binary word (14').

5. A method according to claim 4, wherein the segmentation of the first binary word (12, 12') into the plurality of partial words takes place in an ascending order of the bits of the first binary word (12, 12'); wherein a respective partial word (20) is formed by a number of bits of the first binary word (12, 12') that is equal to the third length (L3); wherein, if the third length (L3) is not a whole number multiple of the first length (L1), the last of the plurality of partial words (20) according to the ascending order is formed by a concatenation of a number of zero-value bits with a residual number of bits of the first binary word (12, 12') so that the sum of the number of zero-value bits and of the residual number of bits of the first binary word (12, 12') is equal to the third length (L3); and / or wherein the second length (L2) and the third length (L3) differ exactly by the number of the at least one information bit (18).

6. A method according to claim 5, wherein the encoding information bit (18) that is concatenated with the last of the plurality of partial words (20) according to the ascending order is set to the first value; and the respective information bit (18) of the remaining partial words (20) is set to the second value.

7. A method according to any one of the preceding claims, wherein a plurality of pixel groups are formed, and indeed in dependence on the similarity of the difference values (24, 24') of the pixels such that the difference values (24, 24') of the pixels of a respective pixel group are more similar to one another than to the difference values (24, 24') of the pixels of a different pixel group.

8. A method according to any one of the preceding claims, wherein, if the first binary word (12, 12') is completely encoded by a plurality of second binary words (14'), the second length (L2) for the plurality of second binary words (14, 14') is constant or differs between the plurality of second binary words (14, 14'); and / or wherein the step of encoding the first binary word (12, 12') by at least one second binary word (14, 14') corresponds to a first encoding method, wherein, for at least some further pixels that are not encoded according to the first encoding method, the first binary word (12, 12') is encoded according to a second encoding method; and / or wherein the image data comprise pixel values of a plurality of mutually different colors, wherein the at least one predetermined other pixel is not arranged directly adjacent to the respective pixel, and wherein the pixel values of the respective pixel and of the at least one predetermined other pixel have the same color; and / or wherein the image data comprise pixel values of a plurality of mutually different colors, wherein the determination of the estimated pixel value of the respective pixel takes place in dependence on the respective pixel value of a plurality of predetermined other pixels, wherein some of these pixels are arranged directly adjacent to the respective pixel and the pixel values thereof have a different color than the pixel value of the respective pixel, and wherein some others of the predetermined other pixels are not arranged directly adjacent to the respective pixel and the pixel values of these other pixels have the same color as the pixel value of the respective pixel.

9. A method according to any one of the preceding claims, wherein the pixels are provided in an arrangement of a plurality of rows, wherein the pixel values are compressed row-wise, and wherein the determination of the estimated pixel value of the respective pixel takes place in dependence on the respective pixel value of a plurality of predetermined other pixels, with the predetermined other pixels being arranged in the row of the respective pixel and / or in an already compressed row, or with the predetermined other pixels only being arranged in the row of the respective pixel; and / or wherein the image data represent a frame from a time sequence of a plurality of frames; and / or wherein the method is configured to compress the image data in a lossless manner; and / or wherein the difference value is quantized according to a difference value quantization rule, with the first binary word representing the quantized difference value; and / or wherein the difference values (24, 24') of a plurality of respective pixels that are each represented by a first binary word (12, 12') are encoded in parallel.

10. A method of compressing and decompressing image data, wherein the image data are compressed in accordance with a method according to any one of the preceding claims and comprise a bitstream (50, 50') having a plurality of bits, wherein the bitstream (50, 50') represents a plurality of compressed pixel values which are associated with a respective pixel, wherein the method for decompressing comprises the following steps: - segmenting the bitstream (50, 50') into a plurality of second binary words (14, 14'), with each of the second binary words (14, 14') having a second length (L2) that is a predetermined length of the respective second binary word (14, 14'), and with each of the second binary words (14, 14') having at least one encoding information bit (18) at a predetermined position in the second binary word (14, 14'); - decoding at least one first binary word (12, 12') by at least one of the second binary words (14, 14'), with the first binary word (12, 12') representing a difference value (24, 24') that corresponds to a predetermined relation between a pixel value of a respective pixel and an estimated pixel value of the respective pixel, wherein it is determined whether the encoding information bit (18) of the at least one of the second binary words (14, 14') is set to a first value, wherein the first binary word (12, 12') is completely decoded by a single one of the second binary words (14) when the encoding information bit (18) of the single one of the second binary words (14) is set to the first value, and wherein the first binary word (12, 12') is otherwise completely decoded by a plurality of the second binary words (14'); - determining a mapped estimated pixel value of the respective pixel in dependence on at least one previously decompressed pixel value that corresponds to the at least one predetermined other pixel using an estimation rule; and - determining a decompressed pixel value of the respective pixel in dependence on the first decoded binary word (12, 12') and on the mapped estimated pixel value of the respective pixel.

11. A method according to claim 10, wherein the first binary word (12, 12') is decoded by at least one of the second binary words (14, 14') in that the encoding information bit (18) is removed from the at least one of the second binary words (14, 14') to form the first binary word (12, 12') or a partial word (20) of the first binary word (12, 12'); and / or wherein a plurality of first binary words (12, 12') are decoded in parallel; and / or wherein the first binary word (12, 12') is decoded by the plurality of second binary words (14, 14') in that the following steps are carried out: - identifying a set of second binary words (14') that corresponds to the plurality of second binary words (14') for decoding the first binary word (12, 12'), wherein the second binary words (14') of the set are identified in that, starting with a respective second binary word (14') of the bitstream (50') whose encoding information bit (18) is not set to the first value and that forms a first element of the set, the respective encoding information bits (18) of the second binary words (14') following according to an ascending order of the bitstream (50') are evaluated to determine whether the respective encoding information bits (18) are set to the first value, with the subsequent second binary words (14') forming further elements of the set when the respective encoding information bits (18) are not set to the first value, with that one of the subsequent second binary words (14') whose encoding information bit (18) is set to the first value forming the last element of the set; - forming a plurality of partial words (20) in that at least the encoding information bit (18) is removed from each of the second binary words (14') of the set; and - forming the first binary word (12, 12') in that the plurality of partial words (20) are concatenated with one another.

12. A data carrier having software stored on the data carrier, comprising commands which, when the software is executed by a computer, cause the computer to carry out the method according to any one of the claims 1 to 9 and / or according to one of the claims 10 to 11.

13. An apparatus having a data carrier according to claim 12 and having a computing device, wherein the computing device is adapted to execute the software stored on the data carrier; wherein the computing device preferably has at least one graphics processor and / or is connected to at least one graphics processor; and / or wherein the apparatus is preferably configured for recording a time sequence of frames, in particular wherein the apparatus is a moving image camera, with the computing device being configured to compress at least one of the frames in accordance with the method according to any one of the claims 1 to 9 and / or to decompress a bitstream that represents at least one of the frames in accordance with the method according to one of the claims 10 to 11.

14. Software on a data carrier comprising commands which, when the software is executed by a computer, cause the computer to carry out the method according to any one of the claims 1 to 9 and / or according to one of the claims 10 to 11.