TW200404268A - Analyzing an image composed of a matrix of pixels - Google Patents

Analyzing an image composed of a matrix of pixels Download PDF

Info

Publication number
TW200404268A
TW200404268A TW092106794A TW92106794A TW200404268A TW 200404268 A TW200404268 A TW 200404268A TW 092106794 A TW092106794 A TW 092106794A TW 92106794 A TW92106794 A TW 92106794A TW 200404268 A TW200404268 A TW 200404268A
Authority
TW
Taiwan
Prior art keywords
classified
pixel
pixels
value
image content
Prior art date
Application number
TW092106794A
Other languages
Chinese (zh)
Inventor
Matteo Marconi
Andrea Rizzi
Mario Trevisian
Original Assignee
Koninkl Philips Electronics Nv
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Koninkl Philips Electronics Nv filed Critical Koninkl Philips Electronics Nv
Publication of TW200404268A publication Critical patent/TW200404268A/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/40Analysis of texture
    • G06T7/41Analysis of texture based on statistical description of texture

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

Method of analyzing an image composed of a matrix of pixels, each pixel being defined by at least one parameter value arranged in a parameter matrix (Y). The method comprises the steps of performing a differentiation operation on the parameter matrix (Y) resulting in a plurality of differentiation entries (D) arranged in a matrix. Sets of neighboring differentiation entries having nearly the same differentiation entry are determined. A path length is determined for the pixels in each set, indicating a number of entries in that set. Based on the path length and differentiation entry, a pixel is classified as natural or synthetic image content.

Description

200404268 玖、發明說明: 【發明所屬之技術領域】 本發明關於分析影像之方法。本發明亦關於一電腦程式 產m,一裝置,一處理器裝置系統及一電腦可讀媒體。 【先前技術】 CRT/LCD監視器,即可應用於電腦及電腦系統之監視器之 特徵為具有高析像度及低亮度。此等顯示系統以下稱為監 視器,典型用於顯不綜合内容,即,正文及圖形。其在目 前有曰增之需求以顯示自然内容,即影像及視頻。與τν— 監視器比較,CRT-監視器之特徵為高析像度及低亮度。此 係因為事實,即原始顯示在監視器上之内容,即電腦之 監視器上,係僅為綜合性,特別由正文所代表。此型内容 非常需要高析像度方能取悅用戶,但亦造成降低亮度,例 如因為CRT中需要之電子束之小光點尺寸。 练合内容以下應瞭解為影像之内容,與自然内容比較, 因其练合來源,而有一較高等級。除以上所提綜合内容之 例外 圖符,符號及任何類別構成之型式,圖形或圖像亦 被w為练合性内容。對比之下,自然内容應瞭解為源自自 、然之任何影像’特別是並非由數位化像片,視頻剪貼及相 似影像,其為任意輸入混合光柵影像之一部分。 目則情勢已有大幅改變··網際網路及多媒體科技(如DVD 及影像儲存及傳輸)已在監視器應用中造成τν-相似内容之 增加。此一新情勢造成監視器之嚴重問題,因為其原始並 非供顯不綜合及自然内容而設計。已對任何供PC應用之監 200404268 此:所面對,特別是CRT-及LCD_視器及平面監視器。此外, 問題已由任何相關處理器裝置系統所遭遇,如視頻 二頻晶片’多媒體晶片,處理器單元,悄緩衝器及VRAM 叫,彼等適於處理包括綜合及自然内容之影像。 下β及自然内容之影像被認為具有混合内容。 :則之-般顯示裝置,特別是監視器裝置,裝置及處理 …置系統必需以高析像度,可忽視之亮度損《,可接受 之處理性能及時間供混合内容實施影像處理。 &已有改進之解決方案可大幅改進性能,特別是視覺性 此為有效改進影像處理性能,該處理可加在及限制在顯 不μ統之特定區域,即,自然内容出現之螢幕。此種由用 戶所選擇之”視窗基”手動選擇為-簡單但有效 一 法,§整個視窗内容為自然時可採納此方法。 义根據美專利號碼6,1 95, 459 Β1所揭示之模糊偵測法則之 影像偵測之偵測方案可以應用,但缺少可靠性。 〜不幸的疋,同一方法不能用於在同一視窗内之混合内 奋典型如網頁,因為改進技術應用在純正文或圖形會造 f視覺品質之大幅損失。用以改進顯示在監視器上^ = 严像之視頻改進算法,當應用在純正文及圖形時有不良影 響。因此,該應用之算法之結果經常不足。 土結果,對任何混合内容之影像,在自然與综合内容間可 罪之區別能力變為非常重要。 【發明内容】 本發月之目的為提供用以分析一影像之改進方法 200404268 即,特別是偵測之區域包括自然影像内容,及該區域包括 混合内容之數位影像之綜合影像内容。此外,本發明之一 目的為提供一電腦程式產品,一裝置,一處理器裝置系統 及電腦可讀取媒體,該媒體可實施該方法。 本發明由附屬申請專利範圍所限定。該附屬申請專利範 圍限定優異之實施例。 本發明之基本優點為,本發明之方法可自動適應影像内 容之影像處理,該内容係在一特殊瞬間顯示或處理。此基 本觀念為自然影像内容可以與綜合影像内容區別。此一區 別係根據本發明之基本觀念由混合局部影像資訊及全局影 像資訊之方式實施。全局資訊為路徑長度有關資訊。局部 資訊係由微分欄為每一像素提供。 可自此種區別獲益之其他應用為影像壓縮技術,該技術 可利用分別編碼方案供自然及綜合影像内容之用。特別 是,此舉係關於加碼視頻影像及正文/圖形之影像壓縮技 術。 上所建議之方法可實質上分為三步驟,每一均含驚人觀 念。 在第一步驟中,該方法可處理像素之一或多個相關參數 值。此外,局部影像資訊係由參數矩陣之微分獲得,而全 局影像資訊係由具有相同微分攔之相鄰像素組獲得。 關於第二步驟,徹底實驗觀察顯示,為自然影像内容發 現長路徑之可能性甚低。該可能性隨微分欄之增加而降 低。反之,此一可能性在綜合影像内容中卻為較高。此一 200404268 觀念係利用此一特性來區別自然影像内容,另一方面,及 此合影像内容中之一綜合影像内容。特別是關於混合影像 内容,此一認知可由觀察而解釋,自然影像内容較綜合影 像内今為’,片斷",因此,自然影像内容之路徑長度之變長 甚為困難。以自然影像内容與綜合影像内容之比較而言, 自然影像之路徑變長更不可能。 結果,方法之第三步驟利用門限函數T(D)實施門限作 業。每組之長度計算如下:如路徑長度及微分欄目之組合超 匕該門限’屬於該組之所有像素被考慮為綜合,該組標為 屬於綜合影像内容。否%,像素被認為自然,並被標為屬 於自然影像内容。 關於梯度值,自然影像内容之門限較高,因為自然影名 之特徵為像素間之較低梯度,故長組更為可能。關於較^ 梯度與較長梯度之W,在較低梯度之案例下,長組更; 能為自然影像内容。反之,具有較高梯度值之門限勢必牵 低’因其為典型綜合影像。 建議之方法可提供屍合内容影像中,自綜合影像内容這 |自然影系内容一改進之品質。 在I異、、、。構中,如微分欄D中之像素微分搁不超過最^ 右則像素被分類為背景影相内容。此-改進係根據, 六民值之像素被單獨考慮,因其代表背景,可屬於综合产 :可屬於自然内各。因此,所有各組之值不超過最4 值時’被標籤為屬於—背景影像内容’不論其路徑長短, 如預定最大之差異值為零’另一改進之結構可達成。此 200404268 一結構可降低處理時間。 如至少一參數值對應像素 &度亦甚為優異。亮度為分 析影像之相關參數,因皇舍冬旦 、匕3衫像冑b x之主要部分及含關 於形狀之資訊,換言之,肉 •-心-偵測所需之資訊。 如像素之微分攔由選擇-描择 %焊—梯度之最大值而定,亦為另一 優點: -亮度之第-梯度作為矩陣中沿像素行之第一方向之像 素位置之函數,及 -亮度之第二梯度作為矩陣中與第一方向垂直之第二方 向之另一行像素位置之含數。 第—及第二梯度可能為低度之正負值,或梯度之絕對值。 如最小攔目值為零亦為—優點。其被發現為—適當值以 分類像素為背景影像内容。 在另一優異構型中,像素被分類為背景組之像素被分類 為自然影像内容,如 -背景組較具有像素被分類為綜合影像内容之相鄰組之 預定數目為少,及 -背景組有—較具有像素被分類為自然像素内容之 組之最小數目, 否則被分類為綜合像素内容。 被標為背景影像内容之像素,利用對標為背景影像内容 實施關係分析將其轉換為綜合影像内或自然影像内容 舉可由實施分析背景影像内容之週圍之影像為之。 如在-組被分類為自然影像内容之影像被;類為 200404268 像内容,則可達成另一改進,如 - 相鄰組具有被分類為綜合影像内容 <彳冢素,及 - 該組之路徑長度低於門限長度。 上述所提之改進步驟在前所提及之步 /μ设實施,以改進 影像内容之偵測甚為重要。 此方法如由-第三部加以補充則為—優點。該第三部 中,具有被標為自然影像内容區域中之不規則被改正。特 別是,分類為综合影像内容之相鄰影像系列如其系列長度 低於最大長度,則被分類為自然影像内容。 另-優異結構提供在具有分類為自然影像内容之影像面 積: ~ 每一像素之飽和參數值加以檢查,及 -如具有飽和參數值高於飽和門限之像素百分比超過門 限百分比,該區域之像素被分類為綜合影像内容。 最後,如為以下結構則亦為優異 產生在被分類為自然影像内容區域中影像之微分欄之 絕對值d, - 產生一直方圖H(d) 一 在零與最大範圍值之間有一絕對值(!之範圍, 一 在具有相同絕對值d之區域中包含數個像素作為 絕對值d之函數 - 在絕對值d處具有峰值,如 - 與直方圖值H(d-1)相近,H(d+1)較H(d)為小,及 - H(d)為在絕對值d及最大範圍值間之絕對值範圍 -11- 200404268 中之最高值,及 在該區之像素被分類為綜合影像内容,如: 絕對值d之最低值其H(d)之峰值超過第一門限距 離,或 ^ 屬於直方圖H(d)之二相鄰學值間之絕對值(d)之差 超過第二門限距離。 微分攔之絕對值之直方圖用爽埒眚八 刀口用术也貫匀類為自然影像内容 之是否為真實之自然。 電腦程式產品之目的,當在雷聪 φ — J田隹冤細上執仃時,由本發明執 w 仃建議之方法之電腦程式產品解決。 特別是,電腦程式產品之改進提供下列門限利用上之偽 像素於位置(i,j ) 計算D(i,j)(梯度值) 计异P (i,j )(路徑長度) 計算T(D(i,j))(門限函數,取自查詢表)200404268 发明 Description of the invention: [Technical field to which the invention belongs] The present invention relates to a method for analyzing images. The invention also relates to a computer program product m, a device, a processor device system, and a computer-readable medium. [Previous technology] CRT / LCD monitors, which can be applied to computers and computer systems, are characterized by high resolution and low brightness. These display systems are hereinafter referred to as monitors and are typically used to display comprehensive content, that is, text and graphics. There is currently an increasing demand to display natural content, namely images and videos. Compared with τν— monitors, CRT-monitors are characterized by high resolution and low brightness. This is because of the fact that the content originally displayed on the monitor, that is, the monitor of the computer, is only comprehensive, and is particularly represented by the text. This type of content requires high resolution to please the user, but it also results in reduced brightness, for example, because of the small spot size of the electron beam required in the CRT. The content of the blending should be understood as the content of the image. Compared with the natural content, it has a higher level because of the source of the blending. In addition to the exceptions to the comprehensive content mentioned above, the symbols, symbols and any type of composition, graphics or images are also referred to as training content. In contrast, natural content should be understood as any image that originates from nature, especially from non-digitized photos, video clips, and similar images, which are part of any input mixed raster image. The situation has changed dramatically. Internet and multimedia technologies (such as DVD and image storage and transmission) have caused an increase in τν-similar content in monitor applications. This new situation poses serious problems for monitors, as their original design was not intended to display uncomprehensive and natural content. Have been supervised for any PC application 200404268 This: Faced, especially CRT- and LCD_viewers and flat monitors. In addition, the problem has been encountered by any relevant processor device system, such as video two-frequency chips' multimedia chips, processor units, quiet buffers and VRAM calls, which are suitable for processing images including integrated and natural content. Images with β and natural content are considered to have mixed content. : Then the general display device, especially the monitor device, device and processing… The system must be equipped with high resolution and negligible brightness loss, acceptable processing performance and time for image processing of mixed content. & Existing improved solutions can greatly improve performance, especially visuality. This is to effectively improve the performance of image processing, which can be added to and restricted to specific areas where the system is unobtrusive, that is, screens where natural content appears. This "windows-based" manual selection by the user is-simple but effective. § This method can be adopted when the entire window content is natural. The detection scheme of image detection according to the fuzzy detection principle disclosed in US Patent No. 6,195,459 B1 can be applied, but it lacks reliability. ~ Unfortunately, the same method cannot be used for mixing in the same window, such as web pages, because the application of improved technology to pure text or graphics will cause a significant loss of visual quality. The video improvement algorithm used to improve the display of ^ = strict image on the monitor has adverse effects when applied to pure text and graphics. Therefore, the results of the applied algorithm are often insufficient. As a result, for any mixed content image, the ability to distinguish between sin and natural content becomes very important. [Summary of the Invention] The purpose of this month is to provide an improved method for analyzing an image. 200404268 That is, in particular, the integrated image content of the detected area includes natural image content and digital image of the area including mixed content. In addition, one object of the present invention is to provide a computer program product, a device, a processor device system, and a computer-readable medium, which can implement the method. The invention is defined by the scope of the accompanying application patents. The scope of this subsidiary application patent defines excellent embodiments. The basic advantage of the present invention is that the method of the present invention can automatically adapt to image processing of image content, which is displayed or processed at a special instant. This basic concept is that natural image content can be distinguished from comprehensive image content. This distinction is implemented according to the basic concept of the present invention by mixing local image information and global image information. Global information is information about path length. Local information is provided for each pixel by a differential column. Other applications that can benefit from this distinction are image compression techniques, which can use separate coding schemes for natural and integrated image content. In particular, this is an image compression technique for coded video images and text / graphics. The method proposed above can be essentially divided into three steps, each of which contains amazing ideas. In a first step, the method can process one or more related parameter values of a pixel. In addition, local image information is obtained from the differentiation of the parameter matrix, while global image information is obtained from adjacent pixel groups with the same differentiation. Regarding the second step, thorough experimental observations show that the probability of finding a long path for natural image content is very low. This probability decreases as the differential column increases. On the contrary, this possibility is higher in comprehensive image content. This concept of 200404268 uses this feature to distinguish natural image content, on the other hand, and one of the composite image content is a comprehensive image content. Especially with regard to the content of mixed images, this cognition can be explained by observation. The content of natural images is ′, fragments " compared with those of integrated images. Therefore, it is very difficult to lengthen the path length of natural image content. In terms of comparing natural image content with comprehensive image content, the longer the path of natural image is, the more impossible. As a result, the third step of the method uses a threshold function T (D) to perform a threshold job. The length of each group is calculated as follows: if the combination of path length and differential column exceeds the threshold, all pixels belonging to the group are considered as integrated, and the group is marked as belonging to the integrated image content. No%, pixels are considered natural and marked as belonging to natural image content. Regarding the gradient value, the threshold for the content of natural images is higher. Because the characteristics of natural shadow names are lower gradients between pixels, long groups are more likely. Regarding the W of the longer gradient and the longer gradient, in the case of the lower gradient, the long group is more; it can be natural image content. Conversely, a threshold with a higher gradient value is bound to be lowered 'because it is a typical comprehensive image. The suggested method can provide an improved quality of the composite image content from the composite image content. In I different ,,,,. In the structure, if the pixel differential in the differential column D does not exceed the rightmost pixel, the pixel is classified as the background image content. This improvement is based on the fact that the pixels of the six people are considered separately because they represent the background and can belong to a comprehensive property: they can belong to nature. Therefore, when the values of all groups do not exceed the maximum of four, ‘labeled as belonging to — background image content’, regardless of the length of the path, if the predetermined maximum difference value is zero, another improved structure can be achieved. This 200404268 structure reduces processing time. For example, at least one parameter value corresponding to the pixel & degree is also excellent. Brightness is a relevant parameter for analyzing the image, due to the main part of Huang Dan Dongdan and the dagger 3 shirt like 胄 b x and the information about shape, in other words, the information required for meat • -heart-detection. If the pixel's differential block is determined by the maximum value of the select-describe% weld-gradient, it is also another advantage:-the -gradient of the brightness as a function of the pixel position along the first direction of the pixel row in the matrix, and The second gradient of brightness is used as the fraction of pixel positions in another row of the matrix in a second direction perpendicular to the first direction. The first and second gradients may be low-level positive or negative values, or absolute values of the gradient. If the minimum blocking value is zero, it is also an advantage. It was found to be—the appropriate value is to classify pixels as background image content. In another excellent configuration, pixels whose pixels are classified as the background group are classified as natural image content, such as-the background group is less than a predetermined number of adjacent groups with pixels classified as the integrated image content, and Yes—Compared with the minimum number of pixels that are classified as natural pixel content, otherwise classified as integrated pixel content. The pixels marked as the background image content can be converted into the integrated image or the natural image content by performing the relationship analysis on the background image content. For example, the surrounding image content can be analyzed. If the images in the -group are classified as natural image content; the category is 200404268 image content, then another improvement can be achieved, such as-the adjacent group has classified as integrated image content < The path length is below the threshold length. The improvement steps mentioned above are implemented in the previously mentioned steps / μ settings to improve the detection of image content is very important. This method is complemented by-part III-advantages. In this third part, irregularities in areas marked as natural image content are corrected. In particular, an adjacent image series classified as comprehensive image content is classified as natural image content if its series length is less than the maximum length. In addition, the excellent structure provides an image area with content classified as natural image content: ~ the saturation parameter value of each pixel is checked, and-if the percentage of pixels with saturation parameter values above the saturation threshold exceeds the threshold percentage, the pixels in the area are Classified as comprehensive video content. Finally, if it has the following structure, it is also excellent to generate the absolute value d of the differential column of the image in the area classified as natural image content,-to generate a histogram H (d)-there is an absolute value between zero and the maximum range value The range of (!, A region containing several pixels in the same absolute value d as a function of the absolute value d-has a peak at the absolute value d, such as-similar to the histogram value H (d-1), H ( d + 1) is smaller than H (d), and-H (d) is the highest value in the absolute value range -11-200404268 between the absolute value d and the maximum range value, and the pixels in the area are classified as Comprehensive image content, such as: the lowest value of the absolute value d whose peak of H (d) exceeds the first threshold distance, or ^ the difference between the absolute values (d) of two adjacent academic values belonging to the histogram H (d) exceeds The second threshold distance. The absolute value of the histogram of the differential block is also used to classify whether the natural image content is true or not. The purpose of a computer program product is to be in Leicong φ — J Tian When the grievance is carried out, the computer program product of the method proposed by the present invention is used to solve the problem. The improvement of the computer program product provides the following thresholds to calculate the D (i, j) (gradient value) using the pseudo pixels on the position (i, j). The difference P (i, j) (path length) is calculated T (D (i, j)) (Threshold function, taken from the lookup table)

如D(i,j) = 0,則S(i ’ ]·)被標為BACK,否則 (如 Ρ(ι ’ j)2T(D(i ’ j)) ’ 則犯,】·)被標為 synt 丄如 P(i ’ j) <τ(D (1,j)),則 s (土,j)被標為 nΑτ 標戴认j用以標籤各矩陣之各欄,s(i,為含每一像素 之影像内容之語意矩陣(自然,综合或背景)。 關於裝置之問題由本發明之裝置解決,該裝置包含電路 及/或電腦程式以分析由像素組成之影像,每一像素由至少 —參數值限定’每-像素之至少—參數值之值係安排在一 -12- 200404268 參數矩陣(Y)内 方法如上所述。 δ處理裝置以實施申請專利範圍第i項之 該裒置可為 ^ w 頌示裝置,監視器,電視機或任 1可匕“;1不15及具有處理電路以處理影像之裝置。 處理器裝置系統及/或電腦可讀媒體之目的,由處理器裝 置糸統及/或具有載於埶杆 取於執仃該方法之電腦程式產品之電腦 可讀媒體解決。 【實施方式】 圖1說明分析影像之建議方 圖1中所用之符號為: 法之主要特性之三個步驟 為4數矩陣包含構成影像之像素矩陣之參數值, D為梯度矩陣包含參數值之微分欄, 〃 p為路徑矩陣,其包含每一像素路徑長度。 S為語意矩陣。每一像素,其包含三標 一,如稍後所述。 一影像係由像素之一矩陣組成。每一像素由至少一參數 值限疋。像素之矩陣之參數值安排在參數矩陣¥中。參數矩 陣通常以數位形式使用。如亮度用為一參數則為一優點。 在第一步驟中,一梯度作業丨在矩陣γ之亮度參數值上實 施,梯度由安排在微分矩陣地中之多個微分攔D提供。 在第二步驟中,相鄰微分攔之各組被辨認出彼此互相差 異不超過一預定最大差異值。每一組,由一路徑發現器2 決定路徑長度,指出在該組中各攔之數目。 另一可簡化必要計算之方法為使預定之最大差異值等於 200404268 零。 在第三步驟中,門限檢查3之實施係由檢查每一像素一微 分搁及該像素路徑長度之組合是否超過預定之門限函數 T(D) 〇 將每一組之路徑長度及微分攔之變數與門限函數T(D)比 較’與該組有關之像素如變數超過門限函數,被分類 為綜合影像内容SYNT,如變數在門限函數T(D)以下,則被 分類為自然影像内容NAT。 圖2說明用於一優異實施例中作為較佳門限函數以^)。門 限函數係微分攔一絕對值d之函數。在自然影像中,像素間 較低之梯度較更為可能,導致在統計上看較低梯度之一較 長之路徑長度。此舉導致圖2所示之較佳門限函數τ(ρ),其 中門限函數T(D)在微分欄之絕對值d增加時降低。在圖 之比較步驟中,微分攔與最小欄值比較。 具有微分攔低於或等於最小欄值之像素被分類為背景影 像内容BACK。分類為背景影像内容BACK之像素可能屬於綜 合影像内容SYNT及自然影像内容NAT,需要進一步處理如以 下所示。 如最小棚值為零則為一優點。 如以上所述各步驟之結果語意矩陣s可由包含每一像素 之一標籤NAT,SYNT或BACK組成。 關於梯度作業,許多算符可能適於實施此任務,但在實 驗測試後發現,該結果與利用不同梯度算符之差異不大。 因此,從計算觀點看,即,最大範數,利用最簡單之範數 200404268 乃係一優點:If D (i, j) = 0, then S (i '] ·) is marked as BACK, otherwise (eg P (ι'j) 2T (D (i'j))' is committed,] ·) is marked Is synt 丄 If P (i 'j) < τ (D (1, j)), then s (soil, j) is labeled as nAτ. Label j is used to label the columns of each matrix, s (i, Is a semantic matrix (natural, integrated or background) containing the image content of each pixel. The problem of the device is solved by the device of the present invention, which includes a circuit and / or a computer program to analyze the image composed of pixels, At least-parameter value limit 'at least per pixel-the parameter value is arranged in a parameter matrix (Y) of -12-200404268 as described above. The δ processing device implements the setting of item i of the scope of patent application It can be a chanting device, monitor, television, or any device; 1 to 15 and a device with a processing circuit to process images. The purpose of the processor device system and / or computer-readable medium is by the processor Device system and / or computer readable medium with computer program product carried on the lever to execute the method. [Embodiment] Figure 1 illustrates The recommended symbol for analyzing the image The symbol used in Figure 1 is: The three steps of the main characteristics of the method are a 4-number matrix containing the parameter values of the pixel matrix that constitutes the image, D is the differential matrix of the gradient matrix containing the parameter values, and 〃 p is the path A matrix that contains the path length of each pixel. S is a semantic matrix. Each pixel contains three standard ones, as described later. An image consists of a matrix of pixels. Each pixel is limited by at least one parameter value.疋. The parameter values of the pixel matrix are arranged in the parameter matrix ¥. The parameter matrix is usually used in digital form. If brightness is used as a parameter, it is an advantage. In the first step, a gradient operation 丨 the brightness parameter of the matrix γ The gradient is provided by a plurality of differential blocks D arranged in the differential matrix ground. In the second step, the groups of adjacent differential blocks are identified to differ from each other by not more than a predetermined maximum difference value. Each group A path finder 2 determines the path length and indicates the number of blocks in the group. Another method that can simplify the necessary calculations is to make the predetermined maximum difference equal to 200404268 zero. In the third step, the implementation of threshold check 3 is performed by checking whether each pixel has a differential hold and whether the combination of the path length of the pixel exceeds a predetermined threshold function T (D). The path length and differential block variables of each group are compared with Comparison of threshold function T (D) 'If the variables related to this group exceed the threshold function, they are classified as comprehensive image content SYNT. If the variables are below the threshold function T (D), they are classified as natural image content NAT. Explain that it is used as a better threshold function in an excellent embodiment. ^). The threshold function is a function that differentiates an absolute value d. In natural images, a lower gradient between pixels is more likely, resulting in a statistical look. One of the lower gradients has a longer path length. This results in the better threshold function τ (ρ) shown in Fig. 2, where the threshold function T (D) decreases as the absolute value d of the differential column increases. In the comparison step of the figure, the differential bar is compared with the minimum column value. Pixels with a differential block below or equal to the minimum column value are classified as the background image content BACK. The pixels classified as the background image content BACK may belong to the comprehensive image content SYNT and the natural image content NAT, and need further processing as shown below. It is an advantage if the minimum shed value is zero. The resulting semantic matrix s of each step can be composed of a label NAT, SYNT or BACK for each pixel. Regarding the gradient operation, many operators may be suitable for this task, but after experimental testing, it was found that the result is not much different from the use of different gradient operators. Therefore, from a computational point of view, that is, the largest norm, using the simplest norm 200404268 is an advantage:

[di dj J 其中D(i,〗)代表具有複數個在行方向中標為丨之微分棚及 在為j之微分攔之微分欄之矩陣D,其中 d〖dJ 代表與行方向及列方向之各攔相對之矩 陣Y之部分微分。 特別是,根據下式[di dj J where D (i,〗) represents a matrix D with a plurality of differential sheds marked with 丨 in the row direction and a differential column of the differential block for j, where d 〖dJ represents the number of rows and columns Differentials of the opposite parts of the matrix Y. In particular, according to the following formula

邱,力=max,{|y(/,力 -}^ 一 1,刀|,昨,力 一 y(,,y. — q } 之公式可以提供。通常’任何_之梯度作業均適合,如 J)=II (r(i, j) -Y{i-1, j)i (Y(iy j) ^Y{u j ^ 1)} I v =,: 其中N為一整數。 當N = 2時,可導致歐幾里德規格: 外"·)=w))2 當^,此導致最大規格,如優異實施例所用者。Qiu, force = max, {| y (/, force-} ^^ 1, knife |, Yesterday, the formula for force_y (,, y. — Q} can be provided. Usually, any gradient operation is suitable, Such as J) = II (r (i, j) -Y {i-1, j) i (Y (iy j) ^ Y {uj ^ 1)} I v = ,: where N is an integer. When N = 2 can lead to Euclidean specifications: outer " ·) = w)) 2 When ^, this results in the largest specifications, such as those used in excellent embodiments.

為每像素,微分欄可储存於一額外記憶體中。但在 :=中’儲存微分攔D之矩陣之額外記憶體並不需要 此H μ憶體中。如每""像素之微分攔已計算 因為:數Ρ存於同"^貞記憶體中’其中儲存對應參數 記憶體。 r小丹便用為此,僅需要-鸟 如方法之第一部分之結果 已指定給每_像素。 三標籤NAT,SYNT或BACK之一 -15- 200404268 具有被分類為背景影像内容BACK像素之各組需加以進一 步處理如圖3之流程圖所示。此等組為影像之均勻區域。通 常,均勻區域代表影像之背景,為此理由,其可能屬於影 像之自然内容及/或綜合内容。例如,山水天空之白區域因 為JPEG壓縮在量子化後可能為均勻,同理,圖表之正文有 一均勻之背景。均勻區域中各組之處理以二步驟實施:後路 徑處理步驟4,隨後以短NAT-路徑處理步驟5跟隨。 已發現二顯著特性必須為屬於自然影像之一均勻區域證 實: 1. 圍繞均勻區域影像之區域應不包括(或包括少數)正 文子符及圖表部分,換言之,其不應包含太多分類 為SYNT像素之各組。 2. 均勻區域必須至少與具有分類為NAT之像素部分相 鄰。否則,無法考慮其為視頻影像之一部分。 在後路徑處理步驟4中以分析語意矩陣S,二特性被證 實。如發現分類為BACK之一組有此特性,該組之像素被轉 換為NAT,否則,其被轉換為SYNT。後路徑處理步驟4之結 果,產生一適應之語意矩陣S1,其中包含分類為NAT或SYNT 之像素。 在一較佳實施例中,以上二特性用來分類具有已分類為 背景影像内容BACK之背景組中之像素。背景組中之像素如 符合以下二特性,則分類為自然影像内容: 1 · 背景組中具有較分類為語意影像内容SYNT像素之相 鄰組之預定數目,及 200404268 2. 背景組具有分類為自然影像内容NAT之相鄰組之最 低數目。 在其他案例中,背景組中之像素被分類為綜合影像内容 ςνΜτ 〇For each pixel, the differential column can be stored in an additional memory. However, the extra memory that stores the matrix of the differential block D in: = does not need this H μ memory. For example, the differential block of each " " pixel has been calculated because: the number P is stored in the same " ^ zheng memory ', where the corresponding parameter memory is stored. Little Dan uses it for this purpose and only needs-birds as the result of the first part of the method has been assigned to every pixel. One of the three labels NAT, SYNT or BACK -15- 200404268 Each group with the BACK pixel classified as the background image content needs to be further processed as shown in the flowchart of FIG. 3. These groups are uniform areas of the image. Usually, the uniform area represents the background of the image, and for this reason, it may belong to the natural content and / or comprehensive content of the image. For example, the white areas of mountains, water, and sky may be uniform after quantization due to JPEG compression. Similarly, the text of the chart has a uniform background. The processing of each group in the uniform area is implemented in two steps: post-path processing step 4 and then followed by short NAT-path processing step 5. It has been found that two significant characteristics must be confirmed for one of the homogeneous areas of natural images: 1. The area surrounding the homogeneous area image should not include (or include a few) text sub-characters and graphic parts, in other words, it should not contain too many classifications as SYNT Groups of pixels. 2. The uniform area must be at least adjacent to the portion of the pixel that is classified as NAT. Otherwise, it cannot be considered as part of the video image. In the post-path processing step 4, the semantics matrix S is analyzed, and the two characteristics are verified. If it is found that one of the groups classified as BACK has this characteristic, the pixels of this group are converted to NAT, otherwise, it is converted to SYNT. As a result of the post-path processing step 4, an adaptive semantic matrix S1 is generated, which contains pixels classified as NAT or SYNT. In a preferred embodiment, the above two characteristics are used to classify pixels in a background group that has been classified as background image content BACK. The pixels in the background group are classified as natural image content if they meet the following two characteristics: 1 · The predetermined number of adjacent groups in the background group that are more classified as semantic image content SYNT pixels, and 200404268 2. The background group has been classified as natural The minimum number of adjacent groups of image content NAT. In other cases, the pixels in the background group are classified as comprehensive image content ςνΜτ 〇

W X X 1 X 上述作業之結果為適應之語意矩陣S1,其包含分類為NAT 或SYNT之像素。 當背景被正確分類,較佳之實施例在短NAT-路徑處理步 驟5中,繼續轉換分類為NAT之各組,其並被隔離及其太短 而無法考慮為自然影像内容。此等組可稱為短NAT路徑,並 被考慮為寄生路徑,因為其常由小圖符造成或由以JPEG壓 縮之綜合影像之一部分造成。在短NAT-路徑處理步驟5中, 在短NAT路徑中之像素被轉換為SYNT。短NAT路徑處理之結 果儲存於一第二適應語意矩陣S2中。 應注意,最後二步驟之順序不能反轉甚為重要。 事實上,自然影像之一小部分常被背景自其餘影像所隔 離。此等路徑可稱為真實短NAT路徑,因其被較佳實施例分 類為屬於自然影像,其為太短但非如短NAT路徑之最大部分 之假偵測。如一較佳實施例在處理BACK部分之前處理短NAT 路徑,則所有短NAT路徑被轉換為SYNT ;因此發生某些真實 短NAT路徑,其為短NAT路徑之子組,亦被轉換為SYNT。此 舉將導致偵測性能之降低。因此,保持正確之步驟順序可 避免不當效應。 如圖3所示,BACK步驟4及短NAT-路徑處理步驟5之後,該 算法產生一輸出即第二適應語意矩陣S2,其僅含二種標 200404268 籤咖及sm。每-像素以此二標藏之一分類。標籤為nat 之像素組代表一輸出表徵碼。此表徵碼常包含在其中及在 邊界之不規則。因此,在—優異實施例中應用不規則降低- 步驟6以降低此等不規則。 - 該第二適應語意矩陣在行及列方向被掃描。此舉係供特 性化以下二情況特徵碼至無特徵碼"渡越及相反之"無特 徵碼至特徵碼”渡越。第一個名詞代表一情況,其中該算法 首先遭遇到屬於特徵碼之一像素,之後遭遇一不屬徵 碼之一像素。第二個名詞代表相反之情況。當"特徵碼至無 # 特徵碼"渡越在一線中發現時,計數器增加,直到一相反渡 越發生或線已完成。隨後計數器被評估;如其值低於某2 門限’所有自最後"特徵碼至無特徵碼"事項之像素均被轉 換為屬於特徵碼之像素。 換言之,被分類為綜合影像内容SYNT之相鄰像素系列, 如該系列之長度低於最大長度’則被分類為自然影像内容 NAT 〇W X X 1 X The result of the above operation is the adaptive semantic matrix S1, which contains pixels classified as NAT or SYNT. When the background is correctly classified, in the short NAT-path processing step 5, the preferred embodiment continues to convert the groups classified as NAT, which are isolated and too short to be considered as natural image content. These groups can be referred to as short NAT paths and are considered parasitic paths because they are often caused by small icons or part of a composite image compressed in JPEG. In the short NAT-path processing step 5, the pixels in the short NAT path are converted to SYNT. The results of the short NAT path processing are stored in a second adaptive semantic matrix S2. It is important to note that the order of the last two steps cannot be reversed. In fact, a small part of the natural image is often separated by the background from the rest. These paths can be referred to as true short NAT paths because they are classified as natural images by the preferred embodiment, which is too short but not as false detection as the largest part of short NAT paths. If a preferred embodiment processes short NAT paths before processing the BACK part, all short NAT paths are converted to SYNT; therefore, some real short NAT paths, which are a subset of short NAT paths, are also converted to SYNT. This will reduce the detection performance. Therefore, maintaining the correct sequence of steps can avoid undue effects. As shown in FIG. 3, after step BACK step 4 and short NAT-path processing step 5, the algorithm generates an output, namely, a second adaptive semantic matrix S2, which contains only two kinds of labels 200404268 signing coffee and sm. Per-pixel is classified by one of these two labels. The pixel group labeled nat represents an output characterization code. This characterization code is often included in irregularities at the boundaries. Therefore, in an excellent embodiment, irregularity reduction is applied-step 6 to reduce these irregularities. -The second adaptive semantic matrix is scanned in the row and column directions. This is to characterize the following two cases: feature code to no feature code " crossover and vice versa " no feature code to feature code " overflight. The first noun represents a situation where the algorithm first encounters a feature The code is one pixel, and then encounters a pixel that is not one of the sign codes. The second noun represents the opposite situation. When " character code to no # character code " is found in the line, the counter increases until one On the contrary, the crossing occurs or the line has been completed. Then the counter is evaluated; if its value is lower than a certain 2 threshold, all pixels from the last " signature to no signature " event are converted to pixels belonging to the signature. In other words, The adjacent pixel series that is classified as the comprehensive image content SYNT. If the length of the series is less than the maximum length, it is classified as the natural image content NAT.

此種型式之處理可消除特徵碼不規則之大部分。 最後實施二測m卜⑽處理之終止,其目的為增加自 、然區域偵測之可信度位準。 圖3所示之第一測試丁1係實施在特徵碼中之像素之彩色 飽和參數上。飽和參數之飽和值Sv在分類為自然之區域中 被逐一像素評估。 每一像素之飽和值Sv可自像素之RGB彩色成分由以下公 式獲得: ν + σ +朗 -18- 200404268 飽和門限值Sv為每 ^ ,久做螞甲飽和值宾於 飽^門限值之像素百分比因而衫。如該像素之百分^ 方、粑和门限值’此特徵碼被考慮為一綜合區域,相關 及路径北標籤為綜合影像内容SYNT。 、This type of processing can eliminate most of the feature code irregularities. Finally, the termination of the second measurement process is implemented, the purpose of which is to increase the confidence level of the natural area detection. The first test D1 shown in Fig. 3 was performed on the color saturation parameters of the pixels in the feature code. The saturation value Sv of the saturation parameter is evaluated pixel by pixel in the area classified as natural. The saturation value Sv of each pixel can be obtained from the RGB color component of the pixel by the following formula: ν + σ + Lang-18- 200404268 The saturation threshold value Sv is every ^, and the pixel with a saturation value that is long enough to be satisfied with the threshold value ^ Percentages are therefore shirts. For example, if the percentage, square, threshold, and threshold value of the pixel are used, this feature code is considered as a comprehensive area, and the correlation and path north labels are comprehensive image content SYNT. ,

★最後’為每—認為自然區域,在微分攔之直方圖上實施 第-測試T2。吾人發現該直方圖必須符合在子然影像時之 二標準。因此’第二測試之目的為證實在制出之區域為 自然影像NAT之標準’並繼續保持測試區域之分類為自然影 像内容NAT’如該區域之直方圖符合此二標準。否則,分類 改變為綜合影像内容SYNT。 在一優異實施例中,在測試下之區域中像素之微分搁之 絕對值d用來產生一直方圖H(d)做為絕對值d之函數。 在另一實施例中,利用微分攔之正及負值。 絕對值d之直方圖H(d)供每一 d值包含數個在測試下區域 中之像素,該區域具有微分欄之值d為絕對值。 絕對值d可自零變化至最大範圍值。最大範圍值對應最大 _ 微分攔’其發生在參數矩陣γ中之相鄰各攔具有最大差異, 、例如,因為自零亮度渡越至最大亮度值,或反之亦然。 直方圖H(d)中之峰值由以下標準限定: 1) H(d)為相對最大,即 H(d) >H(d_l)及 H(d) >H(d+l),· 2) H(d)為範圍(d,最大範圍值)中之絕對最大。 在自然影像範圍NAT之影像直方圖中,與視頻影像相似, 所有峰值(如大於一)出現在低梯度,在峰值間之距離非常 -19- 200404268 小。在正文及圖形影像之直方圖中,峰值間之距離較二視 頻影像大幅增加。在處理數個輸入影像後,第一門限距離 之適s門限值’其為4之最低值之第一 ♦值之絕對值,及相 鄰二峰值間之第二門限距離已被發現。 根據上述二門限值,為保持此分類,自然影像内容NAT 區域必須符合之二標準為: 1) 第峰值必須出現在低於第一門限距離之絕對值d, 及 2) 二連續峰值間之距離必須小於第二門限距離。 參 在一優異實施例中,應用此標準之分類被執行如下: 產生为類為自然影像内容(NAT )之影像之一區域中像 素之微分欄之絕對值d - 產生一直方圖H(d) - 具有在零與最大範圍值之間之絕對值d之範圍, - 在具有相同絕對值d為絕對值d範圍之函數之區域 中包含像素之計數 - 在絕對值d具有峰值,如 鲁 - 相鄰直方圖值H(d-1),H(dH)小於H(d),及 - H(d)在絕對值d與最大範圍值間之絕對值d之範 圍為最T%值,及 - 在該區域中像素被分類為綜合影像内容(SYNT),如: - 絕對值d之最低值之H(d)之峰值超過第一門限距 離,或 - 屬於直方圖H(d)之二相鄰峰值之絕對值(d)間之差 -20- 200404268 超過第二門限距離。 如圖3所示之最後表徵碼Μ,即標籤為NAT像素之區域,其 保持在二測試T1及T2後自然内容NAT分類,以良好之信心位 準代表影像内之一自然區域。 應注意上述實施例僅說明本發明,而非限制。精於此技 藝人士可設許多被選實施例而不致有悖本發明申請專利範 圍之範疇。其中任何置於括弧間之參考符號不應蹶釋為對 申請專利之限制。動詞"包含"及其共軛字不排除請專利範★ The last one is every-think natural area, implement the first-test T2 on the differential histogram. I have found that the histogram must meet the two criteria in Ziran video. Therefore, the purpose of the 'second test is to confirm that the produced area is the standard of natural image NAT' and to keep the test area classified as natural image content NAT 'if the histogram of the area meets these two standards. Otherwise, the classification changes to Synthetic Image Content SYNT. In an excellent embodiment, the absolute value d of the differential of the pixels in the area under test is used to generate a histogram H (d) as a function of the absolute value d. In another embodiment, the positive and negative values of the differential block are used. The histogram H (d) of the absolute value d is for each d value to include several pixels in the area under test, and the value d of the area with the differential column is the absolute value. The absolute value d can be changed from zero to the maximum range value. The maximum range value corresponds to the maximum _ differential block, and its neighboring blocks that occur in the parameter matrix γ have the largest difference, for example, because it crossed from zero brightness to the maximum brightness value, or vice versa. The peaks in the histogram H (d) are defined by the following criteria: 1) H (d) is relatively large, that is, H (d) > H (d_l) and H (d) > H (d + l), · 2) H (d) is the absolute maximum in the range (d, the maximum range value). In the image histogram of the natural image range NAT, similar to the video image, all peaks (such as greater than one) appear at low gradients, and the distance between the peaks is very small -19- 200404268. In the histogram of the text and graphic images, the distance between the peaks is significantly increased compared to the two-video image. After processing several input images, the appropriate threshold value of the first threshold distance, which is the absolute value of the first value of the lowest value of 4, and the second threshold distance between two adjacent peaks have been found. According to the above two thresholds, in order to maintain this classification, the two criteria that the natural image content NAT area must meet are: 1) the first peak must appear below the absolute value d of the first threshold distance, and 2) the distance between two consecutive peaks Must be less than the second threshold distance. In an excellent embodiment, the classification to which this criterion is applied is performed as follows: Generate the absolute value d of the differential column of the pixels in an area of the image that is similar to natural image content (NAT)-generate a histogram H (d) -A range with an absolute value d between zero and the maximum range value,-a count containing pixels in a region having the same absolute value d as a function of the range of the absolute value d-a peak at the absolute value d, such as Lu-phase Adjacent histogram values H (d-1), H (dH) are smaller than H (d), and the range of absolute value d between absolute value d and the maximum range value of H (d) is the highest T% value, and- Pixels in this area are classified as comprehensive image content (SYNT), such as:-the peak value of H (d), the lowest value of the absolute value d, exceeds the first threshold distance, or-belongs to the histogram H (d) two adjacent The difference between the absolute value (d) of the peak value -20-200404268 exceeds the second threshold distance. The final characterization code M shown in Figure 3, that is, the area labeled as NAT pixels, is maintained in the natural content NAT classification after the two tests T1 and T2, and represents a natural area in the image with a good level of confidence. It should be noted that the above-mentioned embodiments are only illustrative of the present invention, but not limiting. Those skilled in the art can design many selected embodiments without departing from the scope of the present invention. Any reference signs placed between parentheses shall not be construed as a limitation on patent application. The verb " include " and its conjugate do not preclude patenting

圍以外之元件或步驟之存在。名詞"一"在—元件之前不排 除有複數個元件。本發明可由包含數元件之硬體及適當程 式之電腦實施。在裝置申請專利範圍中之數裝置,其可合 併於硬體之一及相同項目中。在共同相 ^ " r m ^ 牡〃、U相異之附屬申請專利 &圍中引述之措施不表示此等措施不能 用。 个犯組合一起以利使 【圖式簡單說明】 本發明裝置之各特性將進一 明’其中: 步詳述及配合所附圖式說The presence of elements or steps outside of the scope. The noun " a " does not exclude a plurality of elements before the element. The present invention can be implemented by a computer including hardware and a suitable program including several components. Several devices within the scope of the device patent application can be incorporated in one of the hardware and in the same project. The measures quoted in the common application ^ " r m ^ 〃, U and U are different and do not indicate that these measures cannot be used. Combination of individual offenders to facilitate the use [Simplified illustration of the drawings] Each characteristic of the device of the present invention will be further explained 'Among them: Step by step details and cooperation with the attached drawings

圖1為一流程圖,說明本方法之第一實施例; 圖2顯示第一實施例使用之門限函數; 實施例 圖3為流程圖,說明實施第一實施例之第 圖式代表符號說明 梯度作業 路徑發現器 門限檢查 -21- 200404268 4 BACK路徑處理 5 短NAT路徑處理 6 不正常降低作業 \7 1 參數矩陣 D 梯度矩陣 P 路徑矩陣 S 語意矩陣 NAT 自然影像内容 SYNT綜合影像内容 BACK背景影像内容Fig. 1 is a flowchart illustrating the first embodiment of the method; Fig. 2 shows the threshold function used in the first embodiment; Fig. 3 is a flowchart illustrating the implementation of the first embodiment; Job Path Finder Threshold Check-21- 200404268 4 BACK Path Processing 5 Short NAT Path Processing 6 Abnormally Reduced Operations \ 7 1 Parameter Matrix D Gradient Matrix P Path Matrix S Semantic Matrix NAT Natural Image Content SYNT Comprehensive Image Content BACK Background Image Content

Claims (1)

200404268 拾、申請專利範圍: I =分析像素矩陣組成之影像之方法,各像素由至少 一安排在參數矩陣(Y)中一參數值限定,該方法包括 下列步驟: T、X y丄灵她儆分1卞系,u提供複數々 安排在矩陣中之微分攔(D), ~決定共同相異不超過一預定最大差異值之相葬 微分欄之各組, 決疋各紐之路徑長度,指出該組中相鄰微分攔^ 數目,及分配該路徑長度至該組各像素, 將一像素分類 如分配至該像素之路徑長度及該像素微分术 之組合超過門限函數(KD)),分類為综人马 内容(SYNT),及 。&amp; —:分配至該像素之路徑長度及該像素之微《 f且合低於門限函數_.,分類為&quot; 衫像内容(NAT)。 ” 2·如申請專利範圍第丨項之方法 微㈣⑼中該像素之微分搁不超過最二搁 為老景影像内容(BACK)。 、J刀海 其特徵為該預定 其特徵為至少— 3·如申請專利範圍第1項之方法 大差異值為零。 4.如申請專利範圍第丨項之方法 值對應像素之亮度。 其特徵為—像$ 5·如申請專利範圍第4項之方法 200404268 分欄由選擇二梯度之最大值決定: 亮度之第一梯度在矩陣中沿矩陣中像素之行方向 像素之位置之函數,及 :第一梯度在與第一方向垂直之矩陣中像素 /口另仃之第二方向矩陣中之位置之函數。 6·如申請專利範圍第2項之方法,其特徵為該最小攔之 值為零。 7. 如申請專利範圍第2項之方法’其特徵為在具有分類 為背景影像内容(麗)像素之背景組中之像素被分 類為自然影像内容(NAT),如 月厅、、、且較具有分類為綜合影像内容(§ γ N T )像素 之相鄰組之預定數目為少,及 -背景組具有分類為自然影像内容(NAT)影像之相 鄰組之最小數目, 否則被分類為綜合影像内容(SYNT)。 8. 如申請專利範圍第7項之方法’其特徵為在分類為自 然影像内容(NAT) —組中之像素被分類為綜合影像内 容(SYNT),如 - 該相鄰組具有分類為綜合影像内容(g γ n τ )之像 、素,及 - 該組之路徑長度低於門限長度。 9 ·如申清專利範圍第1項之方法’其特徵為分類為綜合 影像内容(SYNT)之相鄰像素系列,如該系列長度低於 最大長度則被分類為自然影像内容(NAT)。 10·如申請專利範圍第1項之方法,其特徵為具有分類為 200404268 自然影像(NAT)像素之影像之區域中: - 檢查各像素之飽和參數值,及 - 如具有飽和參數值高於飽和門限之像素百八比 值超過門限百分比’該區域之像素被分類為綜合 影像内容(SYNT)。 11 ·如申請專利範圍第1項之方法,其特徵為 - 在分類為自然影像内容(NAT)之影像區域内產生 影像微分攔之絕對值d, - 產生一直方圖H(d) -具有在零及最大範圍值間之絕對值^之範圍, -在具有相同絕對值d之區域中包含像素之數目 之計數作為絕對值d範圍之函數 - 在絕對值d處具有峰值,如 一 相鄰直方圖值H(d-1),H(d+1)小於H(d),及 一 H ( d )在絕對值d與最大範圍值間之絕對值d 之範圍中之最高值,及 影像在該區域被分類為綜合影像内容(SYNT), 如: —絕對值d之最低值其H(d)有一峰值超過第一門 v 限距離,或 屬於直方圖H(d)之二相鄰峰值之絕對值(d)間 之差超過第二門限距離。 種可儲存於電腦可讀媒體之電腦程式產品,包含軟 13體裝置用以執行如申請專利範圍第1項之方法。 •種用以分析像素矩陣組成之影像之裝置,各像素由 200404268 至少一參數值限定,各像素之至少一參數值之值安排 在一參數矩陣(Y)中,該方法包含處理裝置,以執行 申請專利範圍第1項之方法。 14. 一種處理器裝置系統及/或電腦可讀媒體,其具有儲 存於其上之電腦程式,以執行如申請專利範圍第1項 之方法。200404268 Patent application scope: I = Method for analyzing an image composed of a matrix of pixels. Each pixel is defined by at least one parameter value arranged in a parameter matrix (Y). The method includes the following steps: T, X y 丄 灵 她 儆Divided into 1 system, u provides the complex number (D) arranged in the matrix, ~ determines the groups of the intersecting differential columns that are mutually different and do not exceed a predetermined maximum difference value, and determines the path length of each link. The number of adjacent differential blocks in the group, and the path length is assigned to each pixel in the group, and a pixel is classified if the combination of the path length assigned to the pixel and the pixel differential operation exceeds the threshold function (KD)), and is classified as Syndicate Content (SYNT), and. &amp; —: The length of the path allocated to the pixel and the pixel's f <f and the sum is below the threshold function _., classified as &quot; shirt-like content (NAT). "2. If the differential differentiation of the pixel in the method of the patent application item No. 丨 is not more than the second one, it is the old scene image content (BACK). J Daohai is characterized by the predetermined feature is at least-3 · If the method of patent application scope item 1 has a large difference value of zero. 4. If the method of patent application scope item 丨 the method value corresponds to the brightness of the pixel. Its characteristics are-like $ 5 The sub-column is determined by choosing the maximum of the two gradients: the function of the first gradient of brightness in the matrix along the row of pixels in the matrix, and the position of the pixel in the matrix; and The function of the position in the second direction matrix of 仃. 6. If the method of the second item of the patent application scope is characterized by the value of the minimum block being zero. 7. If the method of the second item of the patent application scope 'is characterized by Pixels in the background group with pixels classified as background image content (beautiful) are classified as natural image content (NAT), such as Moon Hall, and, and are more similar to pixels classified as comprehensive image content (§ γ NT). The predetermined number of adjacent groups is small, and the background group has the minimum number of adjacent groups classified as natural image content (NAT) images, otherwise it is classified as comprehensive image content (SYNT). The method 'is characterized in that the pixels in the group are classified as comprehensive image content (SYNT) when classified as natural image content (NAT), such as-the adjacent group has images classified as comprehensive image content (g γ n τ) , Prime, and-The path length of this group is lower than the threshold length. 9 · If the method of claim 1 of the patent scope is' characterized by a series of adjacent pixels classified as comprehensive image content (SYNT), if the series has a low length It is classified as natural image content (NAT) at the maximum length. 10. The method according to item 1 of the scope of patent application is characterized by an area with an image classified as 200404268 natural image (NAT) pixels:-check each pixel Saturation parameter values, and-if the ratio of pixels with saturation parameter values above the saturation threshold exceeds the threshold percentage, the pixels in this area are classified as comprehensive image content (SYNT). The method of claim 1 is characterized by-generating the absolute value d of the image differential block in the image area classified as natural image content (NAT),-generating a histogram H (d)-having a value between zero and the maximum The range of absolute values ^ between range values,-a count containing the number of pixels in a region with the same absolute value d as a function of the range of the absolute value d-has a peak at the absolute value d, such as an adjacent histogram value H ( d-1), H (d + 1) is smaller than H (d), and the highest value of H (d) in the range of the absolute value d between the absolute value d and the maximum range value, and the image is classified in this area Synthetic image content (SYNT), such as:-the lowest value of absolute value d whose H (d) has a peak exceeding the first threshold v-limit distance, or the absolute value (d) of two adjacent peaks that belong to the histogram H (d) ) Exceeds the second threshold distance. A computer program product that can be stored in a computer-readable medium, and includes a software device to perform the method as described in the first patent application scope. A device for analyzing an image composed of a matrix of pixels. Each pixel is defined by at least one parameter value of 200404268. The value of at least one parameter value of each pixel is arranged in a parameter matrix (Y). The method includes a processing device to execute The method of applying for the first item of patent scope. 14. A processor device system and / or a computer-readable medium having a computer program stored thereon to execute a method such as the one in the scope of patent application.
TW092106794A 2002-03-26 2003-03-26 Analyzing an image composed of a matrix of pixels TW200404268A (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
EP02076184 2002-03-26

Publications (1)

Publication Number Publication Date
TW200404268A true TW200404268A (en) 2004-03-16

Family

ID=28051811

Family Applications (1)

Application Number Title Priority Date Filing Date
TW092106794A TW200404268A (en) 2002-03-26 2003-03-26 Analyzing an image composed of a matrix of pixels

Country Status (6)

Country Link
EP (1) EP1490834A2 (en)
JP (1) JP2005521169A (en)
CN (1) CN1656516A (en)
AU (1) AU2003212574A1 (en)
TW (1) TW200404268A (en)
WO (1) WO2003081533A2 (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006087666A1 (en) * 2005-02-16 2006-08-24 Koninklijke Philips Electronics N.V. Method for natural content detection and natural content detector
JP4683294B2 (en) * 2006-03-16 2011-05-18 ソニー株式会社 Image processing apparatus and method, program recording medium, and program
EP2442238A1 (en) 2010-09-29 2012-04-18 Accenture Global Services Limited Processing a reusable graphic in a 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 (en) * 2023-12-08 2024-02-13 中国信息通信研究院 Detection method for depth synthesis information

Also Published As

Publication number Publication date
WO2003081533A3 (en) 2004-08-26
EP1490834A2 (en) 2004-12-29
CN1656516A (en) 2005-08-17
WO2003081533A2 (en) 2003-10-02
AU2003212574A1 (en) 2003-10-08
JP2005521169A (en) 2005-07-14
AU2003212574A8 (en) 2003-10-08

Similar Documents

Publication Publication Date Title
Wu et al. A reversible data hiding method with contrast enhancement for medical images
US8259228B2 (en) Method and apparatus for high quality video motion adaptive edge-directional deinterlacing
US7123769B2 (en) Shot boundary detection
US9679366B2 (en) Guided color grading for extended dynamic range
KR102423710B1 (en) Translucent image watermark detection
US8170361B2 (en) Video window detector
Yalman et al. An image interpolation based reversible data hiding method using R-weighted coding
Wandji et al. Detection of copy-move forgery in digital images based on DCT
TW200404268A (en) Analyzing an image composed of a matrix of pixels
US9013633B2 (en) Displaying data on lower resolution displays
TWI223161B (en) Locally enhancing display information
US6999621B2 (en) Text discrimination method and related apparatus
TW200820136A (en) Video content detector
US20040161152A1 (en) Automatic natural content detection in video information
Pandey et al. Investigating the role of global histogram equalization technique for 99mTechnetium-Methylene diphosphonate bone scan image enhancement
JP2008070860A (en) All-purpose video to which high-degree setting is possible and graphic-measuring device
JP2003018413A (en) Image compressor, image compression method, program code and storage medium
US11670262B2 (en) Method of generating OSD data
Mehta Relation between Entropy and Peak Signal to Noise Ratio in Prediction Error Expansion Considering Region of Interest
CN115035151A (en) Method and device for detecting comb distortion, computer equipment and storage medium
AU767645B2 (en) Method of selecting a video key frame for use in small devices
JP4460359B2 (en) Video signal processing apparatus and video signal processing method
Kong et al. Development of the brightness quality improving system with the video region detection algorithm
Sultana et al. IMPLEMENTATION OF IMAGE STEGANOGRAPHY FOR SECURE DATA COMMUNICATION
WO2003049036A2 (en) Discriminating between synthetic and natural image regions