EP1490834A2 - Analyse d'une image composee d'une matrice de pixels - Google Patents
Analyse d'une image composee d'une matrice de pixelsInfo
- Publication number
- EP1490834A2 EP1490834A2 EP03708399A EP03708399A EP1490834A2 EP 1490834 A2 EP1490834 A2 EP 1490834A2 EP 03708399 A EP03708399 A EP 03708399A EP 03708399 A EP03708399 A EP 03708399A EP 1490834 A2 EP1490834 A2 EP 1490834A2
- Authority
- EP
- European Patent Office
- Prior art keywords
- pixels
- image content
- classified
- value
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
Definitions
- the invention relates to a method of analyzing an image.
- the invention also relates to a computer program product, an apparatus, a processor device system and a computer-readable medium.
- CRT/LCD monitors i.e. monitors applicable for computers and computer systems are characterized by a high resolution and a low brightness.
- These display systems hereinafter termed as monitors, are typically used to display synthetic contents, e. g. text or graphics. They are currently more and more in demand for displaying natural contents, e. g. images or video.
- CRT-monitors are characterized by a high resolution and by a lower brightness. This is due to the fact that originally the content displayed on monitors, e. g. monitors for computers etc., was exclusively synthetic and, in particular, was essentially represented by text. This type of content clearly needs a high resolution to be enjoyable to the user but this also causes a decrease of brightness, for example due to a required small spot size of the electron beam in a CRT.
- Synthetic content is hereinafter to be understood as a content of an image which, compared to a natural content, has a somewhat higher degree of order due to its synthetic origin.
- a text as a synthetic content icons, symbols and any kind of a somewhat constructed form, graphics or pictures have to be understood as a synthetic content.
- a natural content is hereinafter to be understood as any image of a natural origin, in particular not a constructed origin like digitized photos, video clips and similar images, which are part of an arbitrary input composite raster image.
- Images comprising synthetic and natural contents are referred to as having a composite content.
- common display devices in particular monitor devices, devices and processor device systems are required to perform image processing with a high resolution and a negligible loss of brightness and also with an acceptable processing performance and time for images of composite content.
- Detection schemes for image detection based on fuzzy detection rules as described in US 6,195,459 Bl may be applicable, but may have a lack of reliability.
- the invention is defined by the independent claims.
- the dependent claims define advantageous embodiments.
- the basic advantage of the invention is that the method according to the invention makes it possible to automatically adapt image processing to the image content that is displayed or processed at a particular moment.
- the basic idea is that the natural image content can be distinguished from the synthetic image content. The distinction is performed in accordance with the basic idea of the invention by means of a mix of local image information and global image information.
- Global information is information about the sets like the path length.
- Local information is provided by the differentiation entries for each pixel.
- image compression techniques which could use separate encoding schemes for natural and synthetic image content, respectively.
- this concerns image compression techniques for e.g. encoding video images and text/graphics.
- the proposed method as described above can be substantially divided into three steps each comprising a surprising perception.
- the method processes one or more relevant parameter values of the pixels. Furthermore, local image information is given by a differentiation of the parameter matrix, while global image information is given by sets of neighboring pixels having the same differentiation entry.
- the third step of the method performs a threshold operation using a threshold function T(D).
- T(D) The length of each set is computed: if a combination of the path length and differentiation entry exceeds the threshold, all the pixels belonging to the set are considered as synthetic and the set is labeled as belonging to synthetic image content. Otherwise, the pixels are considered as natural and the set is labeled as belonging to natural image content.
- the threshold for natural image contents is higher because natural images are generally characterized by lower gradients between pixels so that long sets are more probable. With regard to a comparison of lower gradients and higher gradients, long sets are therefore more probable for natural image contents in the case of lower gradients. On the contrary, the threshold has to be lower with high gradient values because they are typical of synthetic images.
- the proposed method provides an improved quality of distinguishing natural image content from synthetic image content comprised in an image of composite content.
- a pixel is classified as background image content, if the differentiation entry for that pixel in the differentiation entries D is not more than a minimum entry value.
- the improvement arises from the perception that pixels with a low value are considered separately because they represent backgrounds and could belong either to synthetic image contents or to natural image contents.
- all sets with a value not more than the minimum entry value are labeled as belonging to background image contents, irrespective of the path length.
- a further improved configuration is achieved if the predetermined maximum difference value is zero. This configuration reduces processing times.
- the at least one parameter value corresponds to a luminance of the pixels.
- the luminance is a relevant parameter for analyzing the image, because it contains the major part of the image energy and of the information about shapes, in other words, what is needed for content detection.
- the differentiation entry for a pixel is determined by selecting a maximum value of two gradients: - a first gradient of the luminance as a function of a location of the pixel in the matrix in a first direction along a row of pixels in the matrix, and - a second gradient of the luminance as a function of the location in the matrix in a second direction along another row of pixels in the matrix perpendicular to the first direction.
- the first and second gradients may be the positive or negative value of the gradients or the absolute value of the gradients.
- the minimum entry value is zero. It has been found to be a suitable value for classifying a pixel as background image content. In a further advantageous configuration, pixels in a background set having pixels classified as background image content are classified as natural image content, if
- the background set has less than a predetermined number of neighboring sets having pixels classified as synthetic image content
- the background set has a minimum number of adjacent sets having pixels classified as natural image content, and are otherwise classified as synthetic image content.
- the pixels labeled as background image content are converted either to be synthetic image content or natural image content by performing a relationship analysis of the pixels labeled as background image content. This may be performed by analyzing the surrounding image contents of the background image content.
- a further improvement is achieved if those pixels in a set classified as natural image content are classified as synthetic image content if the neighboring sets have pixels classified as synthetic image content, and the path length of the set is below a threshold length. It is important that this last-mentioned improvement step is executed after the previously mentioned steps to improve correct detection of the image content.
- the method may be complemented by a third part.
- irregularities within areas with mainly pixels labeled as natural image content are corrected.
- a series of adjacent pixels classified as synthetic image content are classified as natural image content if a length of the series is below a maximum length.
- a further advantageous configuration provides that in areas of the image having pixels classified as natural image content: a saturation parameter value is checked for each pixel, and - if a percentage of pixels having a saturation parameter value higher than a saturation threshold value exceeds a threshold percentage, the pixels in that area are classified as synthetic image content.
- H(d) a histogram H(d) is generated having a range of absolute values d between zero and a maximum range value, - containing a count of a number of pixels in that area having the same absolute value d as a function of the range of absolute values d having peaks at an absolute value d if - adjacent histogram values H(d-l), H(d+1) are smaller than H(d), and
- H(d) is the highest value in a range of absolute values d between the absolute value d and the maximum range value
- the pixels in that area are classified as synthetic image content if: a lowest value of absolute values d for which H(d) has a peak exceeds a first threshold distance, or a difference between absolute values (d) belonging to two adjacent peaks of the histogram H(d) exceeds a second threshold distance.
- the histogram of the absolute values of the differentiation entries is used to verify whether areas of the image classified as natural image content are really natural.
- the object regarding the computer program product is solved according to the invention by a computer program product executing the method as proposed, when executed on a computer.
- an improvement of the computer program product provides the following pseudo-code on threshold utilization:
- the labels i and j are used to label the entries of the respective matrix, S(i,j) is a semantic matrix containing the image content (natural, synthetic or background) for each pixel.
- an apparatus comprising circuitry and/or computer programs for analyzing an image composed of a matrix of pixels, each pixel being defined by at least one parameter value, the values of the at least one parameter value for each pixel being arranged in a parameter matrix (Y), the apparatus comprising processing circuitry for carrying out the method of claim 1 as described before.
- the apparatus may be a computer, a display device, monitor, a television or any other product comprising a display device or having processing circuitry for processing images.
- the object regarding the processor device system and/or computer-readable medium is solved by a processor device system and/or a computer readable medium having a computer program product loaded for executing the method as proposed.
- FIG. 1 is a flow chart illustrating a first embodiment of the method
- Fig. 2 shows a threshold function used in the first embodiment
- Fig. 3 is a flow chart illustrating a second embodiment complementing the first embodiment.
- Fig. 1 depicts in three steps the main features regarding the proposed method of analyzing an image.
- the parameter matrix comprising parameter values of a matrix of pixels composing the image
- P the path matrix. It contains the path length for each pixel.
- An image is composed of a matrix of pixels. Each pixel is defined by at least one parameter value.
- the parameter values of the matrix of pixels are arranged in a parameter matrix Y.
- the parameter values are usually available in a digital form. It is advantageous if the luminance is used as a parameter.
- a gradient operation 1 is performed on the luminance parameter values of matrix Y and the gradient is provided by a multitude of differentiation entries arranged in a differentiation matrix D.
- a second step sets of neighboring differentiation entries are identified which mutually deviate by not more than a predetermined maximum difference value. For each set, a path length is determined by a path finder 2 indicating the number of entries in the set.
- a threshold check 3 is performed by checking for each pixel whether a combination of a differentiation entry and the path length of that pixel exceeds a predetermined threshold function T(D).
- the pixels related to the set are classified as synthetic image content SYNT if the variables exceed the threshold function T(D) and are classified as natural image contents NAT if the variables remain below the threshold function T(D).
- Fig. 2 depicts a preferred threshold function T(D) as used in an advantageous embodiment.
- the threshold function is given as function of an absolute value d of a differentiation entry.
- d an absolute value of a differentiation entry.
- Pixels having differentiation entries below or equal to the minimum entry value are classified as background image content BACK.
- pixels classified as background image content BACK may belong to synthetic image content SYNT as well as to natural image content NAT, further processing is required as will be shown hereinafter.
- the semantic matrix S can be composed, containing one of the labels NAT, SYNT or BACK for each pixel.
- the gradient operation many operators may be suitable to perform this task, but it has been found after experimental tests that results do not differ significantly by using different gradient operators. Therefore it is advantageous to use the simplest norm from a computational point of view, i.e. the max-norm:
- D(i, j) denotes the matrix D with the plurality of the differentiation entries labeled i in
- D(i, j) max , ⁇ ⁇ Y(i, j) - Y(i - 1, j) ⁇ , ⁇ Y(i, j) - Y(i, j - 1)
- any kind of gradient operation is suitable like:
- N is an integer.
- a differentiation entry may be stored in an extra memory.
- the extra memory for storing the matrix of differentiation entries D is not necessary.
- the matrix Y may be stored in a frame memory. For each pixel for which the differentiation entry has been calculated, one may store this differentiation entry in the same frame memory in which the corresponding parameter value was stored, because this parameter value is never used any more in the subsequent steps of the method. To do this, one may need only an extra line memory.
- NAT has been assigned to each pixel.
- Sets having pixels classified as background image content BACK values are to be processed further as shown in the flow chart of Fig. 3.
- These sets are uniform areas of the image.
- a uniform area represents the background of an image and for this reason it may belong both to a natural content or/and to a synthetic content of an image.
- white areas of a landscape sky may appear uniform after quantization due to JPEG- compression and in the same way a text of a diagram has a uniform background.
- the processing of the sets in uniform areas is performed in two steps: a back path processing step 4, followed by a short-NAT-Paths processing step 5.
- Two distinctive properties have been identified which have to be verified for a uniform area that belongs to a natural image: 1.
- the area of the image surrounding a uniform area should not include (or should include only a little number of) text characters or graphic parts, in other words, it should not contain too many sets with pixels classified as SYNT.
- the uniform area must be at least partially adjacent to sets having pixels classified as NAT. Otherwise, there is no reason for considering them as part of a video image.
- Two properties are verified in the back path processing step 4 by analyzing the semantic matrix S. If a set with pixels classified as BACK with these properties is found, the pixels in that set are converted to NAT, otherwise they are converted to SYNT.
- an adapted semantic matrix S 1 is generated, containing pixels classified as NAT or SYNT.
- the above two properties are used to classify pixels in a background set having pixels classified as background image content BACK.
- the pixels in the background set are classified as natural image content if the following two properties are met:
- the background set has less than a predetermined number of neighboring sets having pixels classified as synthetic image content SYNT, and
- the background set has a minimum number of adjacent sets having pixels classified as natural image content NAT.
- the pixels in the background set are classified as synthetic image content SYNT.
- the result of the operation mentioned above is the adapted semantic matrix SI, containing pixels classified as NAT or SYNT.
- a preferred embodiment continues to convert, in the short NAT-path processing step 5, sets classified as NAT which are both isolated and too short to be considered as natural image content. These sets may be called short-NAT-paths and are considered as spurious paths, because they often result from little icons or from parts of synthetic images compressed with JPEG.
- the short-NAT-path processing step 5 the pixels in a short NAT path are converted to SYNT.
- the result of the short NAT-path processing is stored in a second adapted semantic matrix S2. It is important to note that the order of the last two steps should not be inverted.
- the algorithm produces, as output, the second adapted semantic matrix S2 that contains only two kinds of labels: NAT and SYNT. Each pixel is classified with one of these two labels.
- the sets of pixels labeled NAT represent an output mask. Such a mask often contains some irregularities both within it and on the boundaries. Therefore, an irregularity reduction operation step 6 is applied in an advantageous embodiment in order to reduce these irregularities.
- the second adapted semantic matrix is scanned both in row and in column direction. In doing so, it is possible to characterize the following two situations: the "mask to no-mask"-transition and the opposite "no-mask to mask”-transition.
- the first term refers to a situation in which the algorithm first encounters a pixel belonging to the mask and afterwards encounters a pixel that does not belong to it.
- the second term refers to the opposite situation.
- series of adjacent pixels classified as synthetic image contents SYNT are classified as natural image content NAT if a length of the series is below a maximum length.
- a first test Tl as shown in Fig. 3 is performed on a color saturation parameter of the pixels in the mask.
- Saturation values Sv of the saturation parameter are evaluated pixel by pixel in areas classified as natural.
- the saturation value Sv for each pixel can be derived from RGB color components of that pixel by the formula:
- a saturation threshold value Sv is determined for each pixel, and a percentage of pixels in the mask whose saturation value is above the saturation threshold value is determined. If the percentage of pixels is above the saturation threshold value, this mask is considered as a synthetic area and the related pixels and paths are labeled as synthetic image content SYNT.
- a second test T2 is performed on a histogram of the differentiation entries for each area recognized as natural. It has been found that such a histogram has to meet two criteria in the case of natural images. Therefore, the objective of the second test is to verify these criteria on areas detected as natural image content NAT, and to keep the classification of tested areas as natural image content NAT, if the histogram of that area meets both criteria. Otherwise, the classification is changed into synthetic image content SYNT.
- absolute values d of differentiation entries of the pixels in the area under test are used to generate a histogram H(d) as a function of the absolute values d.
- positive and negative values of the differentiation entries are used.
- the histogram H(d) of absolute values d contains for each value of d a number of pixels in the area under test having the value d as absolute value of the differentiation entry.
- the absolute value d may vary from zero to a maximum range value.
- the maximum range value corresponds to a maximum differentiation entry which occurs when two adjacent entries in the parameter matrix Y have a maximal difference, for example, due to a transition from zero luminance to a maximum luminance value, or vice versa.
- a peak in the histogram H(d) is defined on the basis of the following criteria: 1) H(d) is a relative maximum, that is H(d) > H(d - 1) and H(d) > H(d + 1); 2) H(d) is the absolute maximum in the range (d, maximum range value).
- the two criteria to be met for areas with natural image content NAT in order to keep this classification are:
- the first peak must appear at an absolute value d below the first threshold distance
- the classification applying the criteria is executed as follows: absolute values d of differentiation entries of pixels in an area of the image classified as natural image content (NAT) are generated, a histogram H(d) is generated
- - adjacent histogram values H(d-l), H(d+1) are smaller than H(d), and - H(d) is the highest value in a range of absolute values d between the absolute value d and the maximum range value, and - the pixels in that area are classified as synthetic image content (SYNT) if: a lowest value of absolute values d for which H(d) has a peak exceeds a first threshold distance, or
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
L'invention concerne un procédé d'analyse d'une image composée d'une matrice de pixels, chaque pixel étant défini par au moins une valeur paramétrique disposée dans une matrice de paramètres (Y). Ledit procédé consiste à exécuter une opération de différenciation sur la matrice de paramètres (Y) résultant en une pluralité d'entrées de différenciation (D) disposées dans une matrice. Des ensembles d'entrées de différenciation voisines ayant approximativement la même entrée de différenciation sont déterminés. Une longueur de chemin est déterminée pour les pixels dans chaque ensemble, indiquant un nombre d'entrées dans l'ensemble concerné. En fonction de la longueur de chemin et de l'entrée de différenciation, un pixel est classé comme contenu d'image naturel ou synthétique.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP03708399A EP1490834A2 (fr) | 2002-03-26 | 2003-03-12 | Analyse d'une image composee d'une matrice de pixels |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP02076184 | 2002-03-26 | ||
EP02076184 | 2002-03-26 | ||
EP03708399A EP1490834A2 (fr) | 2002-03-26 | 2003-03-12 | Analyse d'une image composee d'une matrice de pixels |
PCT/IB2003/001100 WO2003081533A2 (fr) | 2002-03-26 | 2003-03-12 | Analyse d'une image composee d'une matrice de pixels |
Publications (1)
Publication Number | Publication Date |
---|---|
EP1490834A2 true EP1490834A2 (fr) | 2004-12-29 |
Family
ID=28051811
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP03708399A Withdrawn EP1490834A2 (fr) | 2002-03-26 | 2003-03-12 | Analyse d'une image composee d'une matrice de pixels |
Country Status (6)
Country | Link |
---|---|
EP (1) | EP1490834A2 (fr) |
JP (1) | JP2005521169A (fr) |
CN (1) | CN1656516A (fr) |
AU (1) | AU2003212574A1 (fr) |
TW (1) | TW200404268A (fr) |
WO (1) | WO2003081533A2 (fr) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006087666A1 (fr) * | 2005-02-16 | 2006-08-24 | Koninklijke Philips Electronics N.V. | Procede de detection de contenu naturel et detecteur de contenu naturel |
JP4683294B2 (ja) * | 2006-03-16 | 2011-05-18 | ソニー株式会社 | 画像処理装置および方法、プログラム記録媒体、並びにプログラム |
EP2442238A1 (fr) | 2010-09-29 | 2012-04-18 | Accenture Global Services Limited | Traitement d'un graphique réutilisable dans un document |
US11521378B2 (en) | 2020-01-06 | 2022-12-06 | International Business Machines Corporation | Refined searching based on detected object configurations |
US11366624B2 (en) | 2020-03-30 | 2022-06-21 | Kyocera Document Solutions Inc. | Super-resolution convolutional neural network with gradient image detection |
CN117390600B (zh) * | 2023-12-08 | 2024-02-13 | 中国信息通信研究院 | 用于深度合成信息的检测方法 |
-
2003
- 2003-03-12 EP EP03708399A patent/EP1490834A2/fr not_active Withdrawn
- 2003-03-12 WO PCT/IB2003/001100 patent/WO2003081533A2/fr not_active Application Discontinuation
- 2003-03-12 CN CNA038120356A patent/CN1656516A/zh active Pending
- 2003-03-12 AU AU2003212574A patent/AU2003212574A1/en not_active Abandoned
- 2003-03-12 JP JP2003579177A patent/JP2005521169A/ja active Pending
- 2003-03-26 TW TW092106794A patent/TW200404268A/zh unknown
Non-Patent Citations (1)
Title |
---|
See references of WO03081533A2 * |
Also Published As
Publication number | Publication date |
---|---|
WO2003081533A2 (fr) | 2003-10-02 |
TW200404268A (en) | 2004-03-16 |
AU2003212574A1 (en) | 2003-10-08 |
WO2003081533A3 (fr) | 2004-08-26 |
AU2003212574A8 (en) | 2003-10-08 |
CN1656516A (zh) | 2005-08-17 |
JP2005521169A (ja) | 2005-07-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US5073953A (en) | System and method for automatic document segmentation | |
US9092668B2 (en) | Identifying picture areas based on gradient image analysis | |
JP4667062B2 (ja) | 画像解析装置、画像解析方法、及びブロブ識別装置 | |
US6639998B1 (en) | Method of detecting a specific object in an image signal | |
US8787690B2 (en) | Binarizing an image | |
US5745190A (en) | Method and apparatus for supplying data | |
US7639880B2 (en) | Compressing a multivalue image with control of memory space requirement | |
JP2000132690A (ja) | ト―クン化によるイメ―ジ分割を用いたイメ―ジ処理方法および装置 | |
US20100158362A1 (en) | Image processing | |
US20090060354A1 (en) | Reducing Compression Artifacts in Multi-Layer Images | |
US8170361B2 (en) | Video window detector | |
US7075681B1 (en) | System and method for reducing the data volume of images | |
US8023731B2 (en) | Apparatus and method for histogram analysis of image and luminance compensation apparatus using the same | |
EP1490834A2 (fr) | Analyse d'une image composee d'une matrice de pixels | |
US7920755B2 (en) | Video content detector | |
US8472716B2 (en) | Block-based noise detection and reduction method with pixel level classification granularity | |
US20040161152A1 (en) | Automatic natural content detection in video information | |
US20030021476A1 (en) | Text discrimination method and related apparatus | |
US7265873B1 (en) | Image processor that performs edge enhancement using a dynamic threshold | |
JP2004110606A (ja) | 画像処理装置及び方法並びにプログラム | |
US8098945B2 (en) | Method and system for binarizing an image | |
JP2003346081A (ja) | 文字認識装置 | |
GB2434496A (en) | Detection of uniform video frames by summing oppositely signed weighted data value ranges | |
US11295452B1 (en) | Automated method and apparatus for detecting black borders in an image frame | |
AU2007249099B2 (en) | Block-based noise detection and reduction method with pixel level classification granularity |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
AK | Designated contracting states |
Kind code of ref document: A2 Designated state(s): AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LI LU MC NL PT RO SE SI SK TR |
|
AX | Request for extension of the european patent |
Extension state: AL LT LV MK |
|
17P | Request for examination filed |
Effective date: 20050228 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN |
|
18W | Application withdrawn |
Effective date: 20070521 |