1353166 玖、發明說明: · (~ )發明所屬之技術領域 本發明大致揭示一種用於影像處理之的方法及系統,更 明確地,一種用於使得數位掃描文件去除濾選之 〜 刀位及系 統。 (二) 先前技術 幾乎所有印刷品,除了鹵化銀照片之外,使用半色調灑 網(halftone screen)來印刷。這些半色調濾網習用上用於印 刷裝置最佳’而且如果沒有自原掃描影像適當地去除,則 會造成明顯半色調干擾(可見大面積拍差(beating))及可見 之莫爾(Moire)圖型。此種濾網成功地去除而沒有妥協本文 及行項目(line art)品質,是有品質文件掃描及文件分段及 壓縮的基本關鍵。 (三) 發明內容 本發明揭示一種使得影像訊號去除濾選之方法及系統。 該系統包含濾波器組、控制模組及混合模組。濾波器組使 得影像信號來濾波,而產生一組濾波輸出信號。控制模組 根據影像信號及某些濾波輸出信號,來產生至少一個控制 信號。混合模組根據控制信號動態地混合濾波輸出信號, 來產生去除濾選之輸出信號。 (四) 實施方式 在本發明中,說明使得數位掃描文件去除濾選之創新方 法及系統,使得潛在半色調干擾及討厭之莫爾圖型消除或 大幅地減少。本方法使用濾波器組來濾除不同半色調濾選 -5- 1353166 頻率。在其一實施例中,濾波器組之濾波器是在硬體中容 易及有效率地實施之具有可分離金字塔形狀響應的二維濾 波器。滬波器組之輸出以一個像素一個像素方式動態地混 合在一起,而產生去除濾選之輸出。在其一實施例中,本 方法在不同析像度以及頻率及半色調濾網加權量(halftone · weight)度時,使用兩個顏色對比視窗來小心地濾除濾選 · (screen),但是保留本文及行項目邊緣(line art edge)。本 方法也具有功能以使得邊緣銳化來強化本文及行項目,而 且檢測中性(即,無顏色)像素。 φ 在本發明之方法中,重要在注意模糊化(blurring)(低通濾 波)及銳化(sharpening)是獨立地控制。銳化是在模糊化之 後才實施。 本發明之方法可經由使用分段線性控制函數(p i e c e w i s e linear control function)及各種臨界暫存器(threshold register) 來完全地可程式規畫實施。去除濾選截止頻率、半色調濾 網去除之程度及邊緣強化量之選擇可以調整及調諧高品質 輸出。本發明可應用於任何文件掃描產品。 · 本發明之其一實施例以軟體來實施,而且顯示在整個濾 選頻率及印刷大小之大範圍傳送優越影像品質。 第1圖是本發明系統之方塊圖示。系統1 〇 〇包含濾波器 組]〇 2、控制模組1 〇 4及混合模組1 〇 6。 濾波器組1 〇 2接收輸入信號丨〇 !及產生輸入信號經過濾 波版的一組濾波輸出信號1 〇 3,各具有遞增地較大濾波範 圍。在其一實施例中,對已知輸入析像度,所選擇最大濾 -6- 1353166 波器之大小’大約直接地和所要去除濾選之最低頻率的倒 數成比例;而所選擇最小濾波器之大小,大約直接地和所 要去除濾選之最高頻率的倒數成比例。 控制模組1 04接收輸入信號1 0 1及某些濾波輸出信號, 而產生控制信號1 〇 5。控制信號1 〇 5以一個像素—個像素 方式來指示那一臆波輸出信號將要混合在一起及混合比例 。在其一實施例中’控制模組1 〇 4也產生多加控制信號1 〇 7 、1 0 9 ’其分別地就像素中性及邊緣銳化來提供強化控制。 混合模組1 〇 6接收來自濾波器組1 〇 1之濾波輸出信號1 〇 3 及來自控制模組1 〇 4之控制信號1 0 5、] 〇 7、I 0 9。混合模 組1 〇 6根據控制信號1 0 5,來選擇及混合濾波輸出信號。 視需要地,混合模組1 〇 6根據控制信號1 〇 7、1 0 9,也可對 所混合信號來實施邊緣銳化及/或中性處理。混合模組1 0 6 輸出去除濾選之信號1 11。 控制模組1 04同時地不必接收來自濾波器組1 02之信號 ,因爲其等提供到混合模組1 0 6。在其一實施例中,來自 濾波器組1 02之信號在需要時才提供到在控制模組1 04。 第2圖表示本發明系統之其一實施例200。系統200包 含濾波器組202、控制楔組2 0 4及混合模組2〇6。 濾波器組模組202包含5個並聯濾波器,及第6個濾波 器,其串聯連接5個濾波器中最大之一個(就瀘波範圍而言) 。視需要來提供第6濾波器串聯連接5個濾波器中第二大 之一個而不是最大濾波器(如第2圖點線所示)’來降低瀘 波量。 1353166 濾波器組模組2 02包含濾波器陣列B1及濾波器陣列B1 之濾波器中具有最大濾波範圍(即,濾波器2 02 d或2 02 e) 所串列式串聯的濾波器B2。濾波器陣列B1包含濾波器202a ' 202b、202c ' 2 02d、202e。濾波器組模組202在去除濾 選系統200中需要最多計算。 去除濾選系統之目的’在於檢測輸入流中之輸入半色調 ,而選擇性地將其等濾除。主要目的在濾除半色調,而且 保持輸入影像所呈現頁面上之項目(1 i n e a r t)中物件的銳邊 緣。同時,去除濾、選系統可以銳邊緣定義,來視需要地強 化本文或行項目物件,以便不明顯地折衷本文及行項目圖 像(1 i n e a r t g r a p h)之品質。兩作業(濾波及強化)嚴緊地但獨 立地來控制。 去除濾選方法之第一步驟在提供原信號之一組模糊(即 ’低通)信號。去除濾選(去除半色調)作業是以選擇性地一 個像素一個像素方式地混合濾波輸出來獲得。控制邏輯使 用來決定那一輸出將要混合及多少量,因而,自一個像素至 次一像素地提供可變之混合容量(或等效”瞬間頻率控制”)。 雖然通常在任何已知時間不多於兩個濾波輸入信號混合 ’但是,當實際上需要時,其不容易產生所需要這些輸出 。原因在各後續像素會需要混合不同對之濾波輸出信號, 因爲相關之大量則後關係爐波(context filtering ),所以其 會用長時間來產生。更進一步地’一些濾波輸出信號(諸如 在第2圖之這些濾波器202a、202e或可能地202d)隨時也 需要用於去除濾選分析做爲在控制模組2 0 4中之部份檢測 -8- 1353166 及控制邏輯。因此,因爲效率原因,全部濾波輸出信號倂 行地以一個整合模組之濾波器組來產生。所選擇特定濾波 器形狀(三角形且在2D中可分離)能實施先前(較小一號之 大小)濾波器的一個濾波器,因而,大幅地減少計算數量。 濾波器陣列B 1包括5個倂行及獨立之全彩三角形濾波器 ;202a ' 202b、202c' 202d、202e > 其分別是 3x3、5x5' 7x7、9x9、1 lx 1 1之濾波器(分別地表示爲F_3、F_5、F_7 、F_9及F_1 1,而具有下標(index)表示所對應濾波器之大 小)。濾波器陣列B 1如第2圖所示。濾波器陣列B 1之濾波 器是低通爐波器,具有不同截止頻率(cutoff frequency), 便於減少發生在預定範圍內之不同半'色調濾選頻率。在濾 波器陣列中大小最大之濾波器,以所要去除之最低半色調 頻率來決定。因爲現在設計目標定址在達到600dpi掃描, 所以不可能大幅地減小最大濾波器之大小非常地超出其現 在尺寸。 對各濾波器之輸入信號是全彩(L、a、b)源信號(source signal)SRC。在其一實施例中,源信號 SRC之色度通道 (chroma channel)(a、b)僅在快速掃描方向中可以2之因數 來子-抽樣(sub-sample)。24-位元輸入信號SRC進給到濾波 器陣列B 1之全部5個濾波器。濾波器全部以最大輸入資料 速率來作業,各產生獨立全彩濾波輸出,標示爲BLR_n,η 是濾波器範圍(filter span)。 各濾波器獨立地處理(L ' a、b)各彩色分量之輸入資料。 各濾波器是可分離成二個一維濾波器構成的二維濾波器。 -9- 1353166 在其一實施例中,各構成之一維(ID)濾波器具有對稱、三 角形形狀及有整數係數。第3圖所示是濾波器之id離散響 應(discrete response)的實例。 各濾波器輸出規格化回到8 -位元範圍。某些濾波器,諸 如濾波器2 0 2a(F_3濾波器),具有2冪次數之加權總和。 · 這些濾波器在規格化步驟中不需要除算(division),因爲其 ‘ 可容易地以算術移位(arithmetic shift)來實施。例如,F_3 濾波器具有總加權1 + 2 + 1 = 4,而以本加權之除算可簡單算 術右移2位而獲得。 _ 當濾波器之總加權沒有加達2冪次時,密集計算之除算 作業仍可以使用二個數値比例之乘算來近似除算而避免, 其中分母是所選擇2冪次數。 例如,最小濾波器F_3之整個2-D響應是: V *[1 2 1]=丄 •1 2 Γ F 3 =丄 2 2 4 2 - 16 _1_ 16 _1 2 1_ 較大濾波器也可同樣地說明。因爲這濾波器可以分離, 最佳地以相互正交之2個一維步驟來將其實施。爲更有效 率實施,較大濾波器可分享較小濾波器之局部結果,而不 是分別地計算各濾波器之結果。 增加濾波器效率之其一方法,量增加垂直之前後關係 (vertical context)及順序地處理許多行(iine)。例如,最大 濾波器F_1 1需要1 1行輸入來產生單一行輸出(效率〜9%) 。此外,因爲濾波作業進行到頁面下到次一行,所以必需 -10- 1353166 重新讀取1 〇行同時取得新行。濾波器效率以同時地一起處 理更多輸入行來改善。例如,如果輸入行數目1 1增加到 2 0,現在濾波器可產生8行輸出,而效率達到4 0 % = 8 / 2 0。 然而’儲存更多輸入行之較大輸入緩衝器將需要更大管 線延遲。實際上,在濾波器效率及資料頻帶寬相對緩衝記 憶體大小間之適當折衷,視所期望系統成本而定。 濾波器Β2使用來進一步濾出在濾波器陣列Β1之最大濾 波器202 e(F_l 1濾波器)的輸出。濾波器Β2之輸出信號 BLR_ A在去除濾選分析及檢測電路中的某些地方中使用爲 參考信號。因此,信號必需是穩定而且儘可能地沒有雜訊。 濾波器B 2包含如上述在濾波器陣列b 1中所實施最大濾 波器完全相同之F_1 1濾波器。以串列連接濾波器B2及濾 波器202 e ’整個濾波效應等效於以2 2x22濾波器來濾波》 如下文(第8圖)所述,信號BLR_A將不僅使用在混合模 組中之混合’而且用於使得點銳化。 去除濾選頻率範圍如果需要,可以改變濾波器陣列B 1 中之濾波的濾波大小及數量來調整。 控制模組204包含低析像度對比模組B3 '濾波器B4、 半色調預估模組B 6及像素對比模組b 7。 低頻對比模組B 3量測濾波器陣列b 1之最大濾波輸出的 對比量。其接收來自最大濾波器202e(F_ll濾波器)之全彩 濾波信號做爲輸入。輸入信號是24-位元(L ' a、b),其中(a 、b)可在快速掃描方向中以2之因數來子-抽樣。對比模組 B3產生單色輸出(單一通道),其規格化來適合輸出8-位元 -11- 1353166 範圍。對比模組B3使用3個5x5視窗,在有關現有像素上 中心處,每一彩色分量一個視窗。 對比模組之作業如下。對各像素位置,使得5 X 5視窗之 內容搜索最小値及最大値。對各彩色分量之搜索是獨立地 實施。 所組合對比量度定義爲來自各彩色分量之貢獻値平方和 ,即: △ L ~ L m ; a x " L rn i n, (2) △ a: 一 & m a x " & m i n , (3) △ b =b m a x b m i n » (4) 其中(Lr n a x ' L m i n ) ( ^ m a x ' amin)、(bmax、bnijn)疋在個別彩 色分量之5 x5視窗內所發現獨立的最小及最大値,而且輸 出値定義爲 A = (AL2 + Aa2 + Ab2)/C (5) 其中C是比例回復輸出來適合8_位元範圍之規格化常數。 C値是2冪次數値,在於保持設計參數。以規範C値爲2 冪次數値,所需要除算作業可以實施爲簡單算術向右移位 。如果△値變得太大,多加邏輯使用來限制結果値在8-位 元之範圍。 注意輸出對比値是具有和統計差異相同性質之平方和量 度値。其量測在5 x5視窗內側之最大平方對比値。在視窗 內側是否有多於一個像素具有相同最大或最小値不重要, 對比仍然相同。同樣地,如果某些彩色分量在視窗上是常 數,則最大値完全相同於最小値,對比貢獻値將爲零。 -12- 1353166 在某些情形中’尤其對沒有發射許多雜訊之高品質掃描 器’可充分地使用多少小於輸入信號之已濾波版(filtered version)來產生低析像度對比輸出。對這些情形,特別選擇 存在於選擇濾波器組B1之較小F_9濾波器的輸出來替代濾 波器F_ 1 1。本選擇如第2圖點線所示》 低析像度對比濾波器B 4使用來進一步施加在低析像度 對比模組B 3所輸出低析像度對比量測値之進一步濾波。因 爲對穩定信號必需要大程度濾波,所以使用大濾波器F_ 1 1 實施對比濾波器B4。 對比濾波器B 4接收來自低析像度模組B 3之8 ·位元輸出 做爲輸入。其產生濾波及8-位元規格化輸出,表示爲信號 C L 0。爲避免除算做爲部份規格化過程,以兩大數値之比 値做爲非規格化輸出乘算來近似作業就足夠,其中分母是 2冪次數値(以簡單算術移位來實施)。 注意在本實施例中,對比濾波器B 4因而完全相同於濾波 器陣列B 1及濾波器B2中所使用之F_ 1]濾波器,除了其 僅在單一 8-位元彩色分量上作業(相對於B1及B2之全彩 濾波器)。 半色調預估模組使用來量測有關現有像素周圍之小區內 的頻率及半色調加權。對半色調預估模組之輸入是源信號 SRC及來自濾波器陣列B1之最小濾波輸出BLR_3。這兩者 之輸入是全彩(L、a、b)信號.。 半色調預估模組產生兩個單色(單一通道)輸入信號’ FRQ 及HTW,分別地表示預估頻率及半色調加權。各該信號以 1353166 8 -位元方式來表示。HTW表示在半色調之區內的信心度 (level of confidence)。如果HTW小(低信心),則活動分段 (active segmentation)關閉來防止各及每一半色調點之提 升(lifting)。 第4圖是半色調預估模組B6實施例之方塊圖示。 如第4圖所示’半色調預估模組之實施例,包含兩個分 離頻率檢測通道倂行地作業,其輸出在最後步驟結合在一 起而產生半色調加權。各頻率通道包含最小-最大紋理檢測 器,跟隨串列連接之平均濾波器(cascaded averaging filter) 。多數平均濾波器同時也以2之因數來子-抽樣資料(即, 僅產生每一相隔之濾波値)+’使得峰値頻帶寬(peak b a n d w i d t h )大幅地減小,僅在最後,資料向上抽樣回復到 原析像度。 第4圖之符號以描繪各種方塊作爲點線所連接之匹配對 (沒有實際連接)來用於強調在兩頻率通道間之相似性。 方塊之匹配對以相同方塊號碼來表示,具有開頭字母C 用於原頻率通道及D用於模糊頻率通道。因此,匹配對以 (Cn、Dn)來表示,η = [2,···,9]。 使用於各種方塊名稱之符號如下:第一數(下標)表示所 使用視窗之大小;第二數跟著斜線表示在各方向中方塊內 側所實施子-抽樣之量。如此,例如,所標示Β-3/2之濾波 器表示一種模糊濾波器(即,低通濾波器)Β,具有3x3視窗 大小,其輸出在兩個方向中以2之因數來子-抽樣(即,對 每一2x2 = 4輸入像素僅傳送一個輸出)。 1353166 下文更詳細說明在半色調預估模組內所包括之各種方塊。 子-抽樣單元D1以省略輸入之每一相隔像素及行(every other pixel and line)而以4之因數來減少資料速率。子-抽 樣僅需要在模糊頻率通道D內。子-抽樣在全析像度頻率通 道C內不需要。對單元D1之輸入是來自濾波器陣列B1之 最小濾波器的全彩(L、a、b)輸出信號BLR_3。單元D1之 輸出是全彩(L、a、b)信號。因爲單元D1之輸入信號已經 濾波’所以子-抽樣不會造成假信號人爲誤訊(aliasing artifacts) ° 兩個完全相同最小-最大檢測模組C2、D2使用於發現在 輸入信號中之峰値(peak)及谷値(valley)。以計數每單元區 之峰値及谷値的數量,獲得局部頻率之量測度。 各該兩個最小-最大檢測單元接收全彩(L、a、b)信號做 爲輸入。根據下文所說明邏輯,當彩色分量之一的中心像 素相對其鄰接像素是在極値(峰値或谷値)時,各單元使用 3 X 3視窗來表示。 各彩色分量在其自己的3x3視窗內獨立地檢測。自各最 小-最大檢測單元之輸出是2 -位元信號,其表示在極値狀態 之彩色分量總數量。本數量可自零(在極値處沒有彩色分量) 變動到3(全部彩色分量在極値處)。當一個或兩個彩色分量 在極値時,沒有區別那一分量是在極値處;僅輸出在極値 處之分量的總數量。 第5圖表示最小-最大檢測結構。對各彩色分量’首先分 析包圍中心像素(有關現有像素)之8個像素外環(outer ring) •15- 1353166 。8個外環進一步分成二組各4像素如第5圖所示。外環 劃分成兩組有助於在檢測直線分段,做爲半色調中可能_ 誤之警報(因爲最常遇到之半色調通常分類舄叢聚點或行: 濾選)。 本結構圖型之創新特徵,在於配置兩個不同組像素,_ 得對於交叉結構圖型之任何行分段,各該兩組像素 '沒有 $ 全地位在分段之一側上。 對各組,在各組之成分中比較像素値, 來獨立地決定 各組 內 最小及最大値 Ar nax = max(Aij); 全部(i, j )屬於組 ,A ; (6) An nin = min(Aij); 全部(i, j)屬於組 A ; (7) Bn iax = max(Bij); 全部(i,j)屬於組 B ; (8) B„ 1in = min(Bij); 全部(i, j)屬於組 B ; (9) 然後 5 總外環對比由上述値來計算 Arjng = max(Amax - Bmax)fcinin(Amin ' lim) (10)1353166 发明, DESCRIPTION OF THE INVENTION: (~) FIELD OF THE INVENTION The present invention generally discloses a method and system for image processing, and more specifically, a tool position and system for de-filtering digital scanning files . (b) Prior Art Almost all prints, except for silver halide photographs, were printed using a halftone screen. These halftone filters are conventionally used for printing equipment best' and if not properly removed from the original scanned image, it will cause significant halftone interference (visible large area beating) and visible Moire Graphic type. This filter has been successfully removed without compromising the quality of this article and line art, which is the key to quality document scanning and file segmentation and compression. (C) SUMMARY OF THE INVENTION The present invention discloses a method and system for filtering and filtering image signals. The system includes a filter bank, a control module, and a hybrid module. The filter bank filters the image signal to produce a set of filtered output signals. The control module generates at least one control signal based on the image signal and some of the filtered output signals. The hybrid module dynamically mixes the filtered output signals according to the control signals to produce an output signal that removes the filtering. (D) Embodiments In the present invention, an innovative method and system for de-filtering digital scanning files is illustrated, so that potential halftone interference and annoying Moire pattern are eliminated or substantially reduced. This method uses a filter bank to filter out the different halftone filters -5 - 1353166 frequency. In one embodiment, the filter bank filter is a two-dimensional filter having a separable pyramid shape response that is easily and efficiently implemented in hardware. The output of the WB group is dynamically mixed in a pixel by pixel manner, resulting in an output that removes the filter. In one embodiment, the method uses two color contrast windows to carefully filter out the screens at different resolutions and frequency and halftone weights, but Keep this article and line art edge. The method also has the functionality to make edge sharpening to enhance the text and line items, and to detect neutral (ie, no color) pixels. φ In the method of the present invention, it is important to note that blurring (low pass filtering) and sharpening are independently controlled. Sharpening is implemented after the fuzzification. The method of the present invention can be fully programmable by using a piecewise linear control function (p e e e e e s e linear control function) and various threshold registers. The selection of the filter cutoff frequency, the degree of halftone filter removal, and the amount of edge enhancement can be adjusted and tuned for high quality output. The invention is applicable to any document scanning product. An embodiment of the present invention is implemented in software and exhibits superior image quality over a wide range of filter frequencies and print sizes. Figure 1 is a block diagram of the system of the present invention. System 1 〇 〇 contains filter bank 〇 2, control module 1 〇 4 and hybrid module 1 〇 6. Filter bank 1 〇 2 receives the input signal 丨〇 ! and generates a set of filtered output signals 1 〇 3 of the filtered version of the input signal, each having an incrementally larger filtering range. In one embodiment, for a known input resolution, the size of the selected maximum filter -6 - 1353 166 is approximately proportional to the reciprocal of the lowest frequency at which the filter is to be removed; and the selected minimum filter The size is approximately proportional to the reciprocal of the highest frequency at which the filter is to be removed. The control module 104 receives the input signal 1 0 1 and some of the filtered output signals to generate a control signal 1 〇 5. The control signal 1 〇 5 indicates in a pixel-by-pixel manner which chopping output signal is to be mixed and mixed. In one embodiment, the control module 1 〇 4 also generates a plurality of control signals 1 〇 7 , 1 0 9 ' which provide enhanced control for pixel neutrality and edge sharpening, respectively. The hybrid module 1 〇 6 receives the filtered output signal 1 〇 3 from the filter bank 1 及 1 and the control signals 1 0 5, 〇 7, I 0 9 from the control module 1 〇 4 . The mixed mode group 1 〇 6 selects and mixes the filtered output signals according to the control signal 1 0 5 . If desired, the hybrid module 1 〇 6 can also perform edge sharpening and/or neutral processing on the mixed signals based on the control signals 1 〇 7 and 1 0 9. The hybrid module 1 0 6 outputs the filtered filtered signal 1 11 . The control module 104 does not have to receive the signal from the filter bank 102 at the same time because it is supplied to the hybrid module 106. In one embodiment, the signal from filter bank 102 is provided to control module 104 when needed. Figure 2 shows an embodiment 200 of the system of the present invention. System 200 includes a filter bank 202, a control wedge set 220, and a hybrid module 2〇6. The filter bank module 202 includes five parallel filters, and a sixth filter that is connected in series to the largest of the five filters (in terms of the chopping range). If necessary, the sixth filter is connected in series to connect the second largest of the five filters instead of the maximum filter (as shown by the dotted line in Fig. 2) to reduce the amount of ripple. 1353166 The filter bank module 2 02 includes a filter B2 of the filter array B1 and the filter array B1 having the largest filtering range (ie, the filter 2 02 d or 2 02 e) in series. Filter array B1 includes filters 202a '202b, 202c' 02d, 202e. The filter bank module 202 requires the most calculations in the removal filter system 200. The purpose of removing the filtration system is to detect the input halftones in the input stream and selectively filter them out. The main purpose is to filter out the halftones and to maintain the sharp edges of the objects in the item (1 i n e a r t) on the page presented by the input image. At the same time, the removal filter and selection system can be sharp edge defined to enhance the article or line item as needed, so as not to significantly compromise the quality of the image and line item image (1 i n e a r t g r a p h). Both operations (filtering and hardening) are controlled strictly but independently. The first step of removing the filtering method is to provide a set of fuzzy (i.e., 'low pass) signals of the original signal. The de-filtering (removing halftone) job is obtained by selectively mixing the filtered outputs one pixel by one pixel. The control logic is used to determine which amount of output will be mixed and thus, providing a variable mixing capacity (or equivalent "instantaneous frequency control" from one pixel to the next. Although usually no more than two filtered input signals are mixed at any known time 'however, when it is actually needed, it is not easy to produce the required output. The reason is that each subsequent pixel will need to mix different pairs of filtered output signals, because the correlation is large, then the context filtering, so it will take a long time to generate. Further, some of the filtered output signals (such as those of the filters 202a, 202e or possibly 202d in FIG. 2) are also required to be used for removal of the filtering analysis as part of the detection in the control module 220. 8- 1353166 and control logic. Therefore, for efficiency reasons, all filtered output signals are generated in turn by a filter bank of an integrated module. The particular filter shape selected (triangular and separable in 2D) enables one filter of the previous (smaller size one) filter, thus greatly reducing the number of calculations. Filter array B 1 includes five limp and independent full color delta filters; 202a '202b, 202c' 202d, 202e > respectively 3x3, 5x5' 7x7, 9x9, 1 lx 1 1 filters (respectively The ground is represented as F_3, F_5, F_7, F_9, and F_1 1, and has a subscript (index) indicating the size of the corresponding filter). The filter array B 1 is as shown in Fig. 2. The filter of filter array B 1 is a low pass filter with different cutoff frequencies to facilitate reducing the different half-tone filter frequencies occurring within a predetermined range. The filter of the largest size in the filter array is determined by the lowest halftone frequency to be removed. Since the design target is now addressed at 600 dpi scans, it is not possible to significantly reduce the size of the maximum filter beyond its current size. The input signal to each filter is a full color (L, a, b) source signal SRC. In one embodiment, the chroma channel (a, b) of the source signal SRC can be sub-sampled only by a factor of two in the fast scan direction. The 24-bit input signal SRC is fed to all five of the filters of the filter array B1. The filters all operate at the maximum input data rate, each producing an independent full color filtered output, labeled BLR_n, where η is the filter span. Each filter independently processes (L ' a, b) the input data of each color component. Each filter is a two-dimensional filter that can be separated into two one-dimensional filters. -9- 1353166 In one embodiment, each of the constituent one-dimensional (ID) filters has a symmetrical, triangular shape and has integer coefficients. Figure 3 shows an example of the filter's id discrete response. Each filter output is normalized back to the 8-bit range. Some filters, such as filter 2 0 2a (F_3 filter), have a weighted sum of two powers. • These filters do not require division in the normalization step because they can be easily implemented with an arithmetic shift. For example, the F_3 filter has a total weight of 1 + 2 + 1 = 4, and the division by this weight can be obtained by simply shifting the right bit by 2 bits. _ When the total weight of the filter is not increased by 2 powers, the calculation of the intensive calculation can still be avoided by using the multiplication of two numbers of scales, where the denominator is the selected number of powers. For example, the entire 2-D response of the minimum filter F_3 is: V *[1 2 1]=丄•1 2 Γ F 3 =丄2 2 4 2 - 16 _1_ 16 _1 2 1_ Large filter can be similarly Description. Since this filter can be separated, it is optimally implemented in two one-dimensional steps that are orthogonal to each other. For more efficient implementations, larger filters can share the partial results of smaller filters rather than calculating the results of each filter separately. One way to increase filter efficiency is to increase the vertical context and to process many rows sequentially. For example, the maximum filter F_1 1 requires 11 input lines to produce a single line output (efficiency ~ 9%). In addition, because the filtering job proceeds to the next line of the page, it is necessary to re-read 1 line and get a new line at the same time. Filter efficiency is improved by processing more input lines together at the same time. For example, if the number of input lines 1 1 is increased to 2 0, the filter now produces 8 lines of output with an efficiency of 40 % = 8 / 2 0. However, larger input buffers that store more input lines will require more pipe delays. In practice, the appropriate trade-off between filter efficiency and data bandwidth versus buffer size depends on the desired system cost. Filter Β 2 is used to further filter out the output of the largest filter 202 e (F_1 filter) at filter array Β1. The output signal of filter Β2, BLR_A, is used as a reference signal in some places in the filter removal analysis and detection circuit. Therefore, the signal must be stable and as free of noise as possible. Filter B 2 contains the F_1 1 filters of the same maximum filter implemented in filter array b 1 as described above. Connecting the filter B2 and the filter 202 e 'in series is equivalent to filtering with a 2 2x22 filter. As described below (Fig. 8), the signal BLR_A will not only be used in the hybrid module' Also used to sharpen the point. The filter frequency range is removed. If necessary, the filter size and number of filters in filter array B 1 can be changed to adjust. The control module 204 includes a low resolution comparison module B3 'filter B4, a halftone estimation module B 6 and a pixel contrast module b 7 . The low frequency comparison module B 3 measures the contrast of the maximum filtered output of the filter array b 1 . It receives the full color filtered signal from the maximum filter 202e (F_ll filter) as an input. The input signal is 24-bit (L ' a, b), where (a , b) can be sub-sampled by a factor of 2 in the fast scan direction. The comparison module B3 produces a monochrome output (single channel) that is normalized to fit the 8-bit -11- 1353166 range. The contrast module B3 uses three 5x5 windows, one for each color component at the center of the existing pixel. The operation of the comparison module is as follows. For each pixel position, the content search of the 5 X 5 window is minimized and maximum. The search for each color component is performed independently. The combined contrast measure is defined as the sum of squared contributions from the respective color components, namely: Δ L ~ L m ; ax " L rn in, (2) Δ a: a & max "& min , (3) △ b =bmaxbmin » (4) where (Lr nax ' L min ) ( ^ max ' amin ) , ( bmax , bnijn ) 独立 independent minimum and maximum 发现 found in 5 x 5 windows of individual color components, and output 値Defined as A = (AL2 + Aa2 + Ab2) / C (5) where C is the proportional recovery output to fit the normalization constant of the 8_bit range. C値 is a power of two, which is to maintain the design parameters. With the specification C値 as a power of 2, the required division operation can be implemented as a simple arithmetic shift to the right. If Δ値 becomes too large, more logic is used to limit the result to the 8-bit range. Note that the output contrast 値 is the sum of squares and measures of the same nature as the statistical difference. It measures the maximum squared contrast 内侧 on the inside of the 5 x 5 window. It is not important to have more than one pixel on the inside of the window with the same maximum or minimum, and the contrast is still the same. Similarly, if some of the color components are constant on the window, the maximum 値 is exactly the same as the minimum 値, and the contrast contribution 値 will be zero. -12- 1353166 In some cases 'especially for a high quality scanner that does not emit a lot of noise' can use a filtered version that is less than the input signal to produce a low resolution contrast output. For these cases, the output of the smaller F_9 filter present in the selection filter bank B1 is specifically selected in place of the filter F_1 1 . This selection is as shown by the dotted line in Fig. 2, and the low resolution contrast filter B 4 is used to further apply the further filtering of the low resolution contrast measurement output by the low resolution comparison module B 3 . Since a large degree of filtering is necessary for the stable signal, the contrast filter B4 is implemented using the large filter F_1 1 . The contrast filter B 4 receives the 8 bit output from the low resolution module B 3 as an input. It produces a filtered and 8-bit normalized output, denoted as signal C L 0. In order to avoid the division as a partial normalization process, it is sufficient to approximate the operation by multiplying the ratio of the two numbers as the non-normalized output multiplication, where the denominator is a power of two times (implemented by a simple arithmetic shift). Note that in the present embodiment, the contrast filter B 4 is thus identical to the F_1] filter used in the filter array B 1 and the filter B2 except that it operates only on a single 8-bit color component (relative Full color filter for B1 and B2). The halftone estimation module is used to measure frequency and halftone weighting within cells surrounding existing pixels. The input to the halftone estimation module is the source signal SRC and the minimum filtered output BLR_3 from the filter array B1. The inputs to both are full color (L, a, b) signals. The halftone estimation module produces two monochrome (single channel) input signals 'FRQ and HTW, which represent the estimated frequency and halftone weight, respectively. Each of the signals is represented by a 1353166 8-bit method. HTW indicates the level of confidence in the halftone region. If the HTW is small (low confidence), the active segmentation is turned off to prevent lifting of each and every halftone point. Figure 4 is a block diagram of an embodiment of a halftone estimation module B6. As shown in Fig. 4, the embodiment of the 'halftone estimation module' includes two separate frequency detection channels for the operation of the line, the output of which is combined in the final step to produce halftone weighting. Each frequency channel contains a minimum-maximum texture detector followed by a cascaded averaging filter. Most averaging filters also sub-sampled data by a factor of 2 (ie, only generate each separated filter 値) + 'so that the peak bandwidth is greatly reduced, and only at the end, the data is upsampled. Revert to the original resolution. The symbols in Figure 4 depict the various pairs of squares as matching pairs (without actual connections) to emphasize the similarity between the two frequency channels. The matching pairs of squares are represented by the same block number, with the initial letter C for the original frequency channel and D for the fuzzy frequency channel. Therefore, the matching pair is represented by (Cn, Dn), η = [2, ···, 9]. The symbols used for the various block names are as follows: the first number (subscript) indicates the size of the window used; the second number followed by the slash indicates the amount of sub-sampling implemented in the inner side of the block in each direction. Thus, for example, the filter labeled Β-3/2 represents a fuzzy filter (ie, low-pass filter) Β having a 3x3 window size, the output of which is sub-sampled by a factor of 2 in both directions ( That is, only one output is transmitted for each 2x2 = 4 input pixel). 1353166 The various blocks included in the halftone estimation module are described in more detail below. The sub-sampling unit D1 reduces the data rate by a factor of 4 by omitting every other pixel and line of the input. The sub-sampling only needs to be in the fuzzy frequency channel D. Sub-sampling is not required in full resolution frequency channel C. The input to unit D1 is the full color (L, a, b) output signal BLR_3 from the minimum filter of filter array B1. The output of unit D1 is a full color (L, a, b) signal. Since the input signal of unit D1 has been filtered' so sub-sampling does not cause aliasing artifacts. ° Two identical minimum-maximum detection modules C2, D2 are used to find the peak in the input signal. (peak) and valley (valley). A measure of the local frequency is obtained by counting the number of peaks and valleys per unit area. Each of the two min-max detection units receives a full color (L, a, b) signal as an input. According to the logic described below, when the central pixel of one of the color components is at a pole (peak or valley) relative to its neighboring pixels, each unit is represented by a 3 X 3 window. Each color component is independently detected within its own 3x3 window. The output from each of the minimum-maximum detection units is a 2-bit signal representing the total number of color components in the extreme state. This number can be changed from zero (no color component at the pole) to 3 (all color components are at the extreme). When one or two color components are at the extremes, there is no difference that the component is at the pole; only the total number of components at the pole is output. Figure 5 shows the minimum-maximum detection structure. For each color component', first analyze the 8 pixel outer rings (15-1353166) surrounding the center pixel (about the existing pixels). The 8 outer rings are further divided into two groups of 4 pixels as shown in Fig. 5. Dividing the outer ring into two groups helps detect straight-line segments as an alarm that may be erroneous in halftones (because the most commonly encountered halftones are usually classified as clusters or rows: filtered). An innovative feature of this structural pattern is the configuration of two different sets of pixels, _ for any line segmentation of the cross-structured pattern, each of the two sets of pixels 'not having a full position on one side of the segment. For each group, the pixel 比较 is compared among the components of each group to independently determine the minimum and maximum 値Ar nax = max(Aij) in each group; all (i, j) belong to the group, A; (6) An nin = Min(Aij); all (i, j) belong to group A; (7) Bn iax = max(Bij); all (i, j) belong to group B; (8) B„ 1in = min(Bij); all ( i, j) belongs to group B; (9) Then 5 total outer ring is compared by the above 値 to calculate Arjng = max(Amax - Bmax)fcinin(Amin ' lim) (10)
其次測g式△ Η II g値來看外環是否有任何對比。不管中心像 素値’輸出設定在零(不是極値點)’如果△ H n g之値小於或 等於預定小臨界値T2 : 如果T2)回到(0) ; (1 1} 另一方面,如果在外環有足夠活動(activity)(如外環對比 >T2所示),則實施兩個測試來看中心像素値相對外環値是 否在極値處。中心像素値定義爲在峰値,如果其明顯地大 於任一組之最大像素値: 如果[(Amax + S<X)及(Bmax 各 X)],回到(】);(12) 1353166 其中S是外環對比,以對比定標參數C(contrast scaling parameter)來比例: S = Aring/C ; (13) 在其一實施例中,對比定標參數C設定等於8。定標參 數C之實際値是在輸入之信號雜訊比的函數。期望保持C 値爲最小-最大檢測單元之類屬參數(generic parameter)。C 値可限制在2冪次數,使得其可以以算術移位來實施,節 省實施每像素之除算作業的需要。 同樣地,中心像素値X定義在各値處,如果其顯然地小 φ 於任一組A或8之最小像素値; 如果[(Amin>X + S)及(Bming X)],回到(1) ; (14) 方程式(12)及(14)決定兩個條件,其中來自3x3檢測視窗 之輸出設定爲1 ;在全部其他情形中,輸出設定在0。 在第二實施例中,中心像素値X定義在峰値處,如果其 明顯地大於任一組之最大像素値: 如果[(Α„^χ+ΝΤΗ<Χ)及(Bmax ‘ X)],回到(1) ; (1 2A) 其中Nth是雜訊臨界値,定義爲: φNext, measure the g formula △ Η II g値 to see if there is any contrast in the outer ring. Regardless of the center pixel 値 'output set at zero (not the extreme point) 'if △ H ng is less than or equal to the predetermined small threshold 値 T2 : If T2) returns to (0); (1 1} on the other hand, if If the outer loop has enough activity (as shown by outer loop contrast > T2), then two tests are performed to see if the center pixel is relative to the outer loop. The center pixel is defined as the peak, if It is significantly larger than the maximum pixel of any group: If [(Amax + S<X) and (Bmax each X)], return to (]); (12) 1353166 where S is the outer ring contrast, to compare the calibration Parameter C (contrast scaling parameter) to scale: S = Aring / C; (13) In one embodiment, the comparison calibration parameter C is set equal to 8. The actual 値 of the calibration parameter C is the signal noise ratio at the input. The function is expected to keep C 値 as the minimum-maximum detection unit's generic parameter. C 値 can be limited to 2 powers so that it can be implemented with arithmetic shifts, saving the need to implement a per-pixel divide operation. Similarly, the center pixel 値X is defined at each corner, if it is obvious Small φ is the minimum pixel 任一 of any group A or 8; if [(Amin>X + S) and (Bming X)], return to (1); (14) Equations (12) and (14) determine two Condition, wherein the output from the 3x3 detection window is set to 1; in all other cases, the output is set to 0. In the second embodiment, the central pixel 値X is defined at the peak, if it is significantly larger than either group Maximum pixel 値: If [(Α„^χ+ΝΤΗ<Χ) and (Bmax 'X)], return to (1) ; (1 2A) where Nth is the noise threshold 値, defined as: φ
Nth =雜訊偏離値+(雜訊因數*χ)/256 其中雜訊偏離値(NoiseBias)及雜訊因數(noise factor)是調 整參數(t u n i n g p a r a m e t e r )。 同樣地’中心像素値X定義在各値處,如果其明顯地小 於任一組A或B之最小像素値: 如果[(Amin>X + NTH)及(Bmin 2 X)],回到(1 ) ; ( 1 4 A) 方程式(I2A)及(MA)決定兩個條件,其中來自3x3檢測 -17- 1353166 視窗之輸出設定在1 ;在全部其他情形中,輸出設定 。注意在第二實施例中,不需計算全部外環對比値。 最後,如上所述,各彩色分量獨立地經由其自己的 視窗3 X 3來處理。然後,彩色分量之三個二進位輸出 —起來形成最小-最大檢測模組之最後2 -位元輸出。 兩個最小-最大檢測輸出.C 2及D 2分別地進給到串歹ϋ 之濾波器鏈(cascaded filter chain)C3-C6 及 D3-D6。在 濾波器鏈間,第一濾波單元C3及D3不同,此外,其 續單元C4-C6及D4-D6完全相同。在其一實施例中, 濾波單元是對稱、三角形及可分離之濾波器,如第3 示之形狀相同。 第一濾波器單元C 3接收來自高析像度最小-最大檢 元C2之2-位元輸出。輸入經由如上在其一實施例所翅 對稱、三角形、可分離之濾波器來濾波。濾波器之形 第3圖所示濾波器F_7之形狀相同。 濾波器F_7/4不同於濾波器F_7,在於輸出在兩方向 4之因數來子·抽樣,如以Μ符號所示。意即,濾波器 對於每4個輸入像素及每一個第四行才產生一個輸出 ,因而,有效地以1 6之因數來減少資料頻帶寬。 其他不同於濾波器組之濾波器在於使用不同規格化 。因爲對第一濾波單元C 3之輸入限制在2 -位元(而不 在濾波器組之8·位元輸入),濾波器之輸出以不同之2 數(即2 )定標結果來規格化。定標幕次必需保持設計 (d e s i g n p a r a m e t e r)。然而,在第一次規格化之後,結 在〇 分離 加在 連接 兩個 他後 全部 圖所 測單 7x7 狀和 中以 F_7/4 像素 因數 是如 冪次 參數 果已 1353166 定標來適合8 -位元之範圍,使得自本點以後,後續濾波使 用 8-位兀呈現系統(8-bit representation system)。 第二鏈前置濾波器單元D3不同於C3在兩方面。首先, F_5/2在各方向中僅以2之因數(而不是4)來子-抽樣輸入。 意即’濾波器僅對於每一相隔輸入之像素及每一相隔之行 來產生一個輸出像素,因而,有效地以4之因數來減少資 料頻帶寬。因爲子-抽樣因數較小,所以濾波器之範圍結果 地可自7(對C3)減少到5(對D3)。在其一實施例中,F__5/2 之規格化因數決定爲29。注意來自兩前置濾波器單元C3 反D3之(現在8 -位元)輸出兩者在相同析像度-在兩方向中 以4或原輸入頻帶寬之16分之一來子-抽樣。這是因爲在 上鏈之C3單元的F_7/4濾波器以4來子-抽樣資料,而在 下鏈之55/2及F_5/2單元D3的組合獲得整個輸出速率匹 配C3輸出速率。 來自濾波單元C3及D3之兩個輸出進一步分別地經由三 個多加及完全相同的單元C4-C6及D4-D6來濾波。各該6 個濾波器單元也在兩個方向中以2之因數來子-抽樣資料的 F_3/2濾波器(具有係數1-2-1)來濾波其個別的輸入信號。 注意各該濾波器具有1 +2+ 1 =4之總加權,因而,可以簡單 算術向右移位2來替代規格化除算來簡化實施作業。 因爲各濾波單元也以2之因數來子·抽樣其個別輸入信號 ,所以在C 6及D 6濾波器單元之個別輸出處的信號,在各 方向中實際上以32之因數來子-抽樣(或在頻帶寬中減少 1 0 2 4 次)〇 -19- 1353166 其次兩個濾波器單元C7及D7是特定濾波器單元,以 Fz_5來表示。Z後綴(suffix)表示消除來自總規格化加權之 任何零項的濾波器。通用FZ_n濾波器方程式如下所示:Nth = noise deviation 値 + (noise factor * χ) / 256 where NoiseBias and noise factor are adjustment parameters (t u n i n g p a r a m e t e r ). Similarly, the 'central pixel 値X is defined at each , if it is significantly smaller than the minimum pixel of any group A or B 如果: If [(Amin>X + NTH) and (Bmin 2 X)], return to (1 ( 1 4 A) Equations (I2A) and (MA) determine two conditions, where the output from the 3x3 detection -17- 1353166 window is set to 1; in all other cases, the output is set. Note that in the second embodiment, it is not necessary to calculate the total outer ring contrast 値. Finally, as described above, each color component is processed independently via its own window 3 X 3 . The three binary outputs of the color component are then combined to form the last 2-bit output of the min-max detection module. The two minimum-maximum detection outputs, C2 and D2, are fed to the cascaded filter chains C3-C6 and D3-D6, respectively. The first filtering units C3 and D3 are different between the filter chains, and the remaining units C4-C6 and D4-D6 are identical. In an embodiment thereof, the filtering unit is a symmetric, triangular, and separable filter, and the shape is the same as shown in FIG. The first filter unit C 3 receives the 2-bit output from the high resolution minimum-maximum element C2. The input is filtered via a fin-symmetrical, triangular, separable filter as described above in one embodiment. Shape of the filter The shape of the filter F_7 shown in Fig. 3 is the same. The filter F_7/4 is different from the filter F_7 in that the output is factored in two directions 4, as shown by the Μ symbol. That is, the filter produces an output for every 4 input pixels and every fourth line, thus effectively reducing the data bandwidth by a factor of 16. Other filters that differ from the filter bank are based on different normalizations. Since the input to the first filtering unit C 3 is limited to 2 - bit (instead of the 8 bit input of the filter bank), the output of the filter is normalized by a different number of 2 (i.e., 2) scaling results. The calibration screen must maintain the design (d e s i g n p a r a m e t e r). However, after the first normalization, the knot is added to the 〇 加 加 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 连接 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 The range of bits is such that after this point, the subsequent filtering uses an 8-bit representation system. The second chain pre-filter unit D3 is different from C3 in two respects. First, F_5/2 is sub-sampled in only 2 factors (instead of 4) in each direction. That is, the filter produces an output pixel only for each pixel that is separated from each other and every other row, thus effectively reducing the bandwidth of the data by a factor of four. Since the sub-sampling factor is small, the range of the filter can be reduced from 7 (for C3) to 5 (for D3). In an embodiment thereof, the normalization factor of F__5/2 is determined to be 29. Note that both (now 8-bit) outputs from the two pre-filter units C3 and D3 are at the same resolution - sub-sampled in either of the two directions or one-sixteenth of the original input frequency bandwidth. This is because the F_7/4 filter in the C3 unit of the winding has 4 sub-sampled data, and the combination of 55/2 and F_5/2 unit D3 in the lower chain obtains the entire output rate matching C3 output rate. The two outputs from filtering units C3 and D3 are further filtered separately via three more identical and identical units C4-C6 and D4-D6. Each of the six filter units also filters its individual input signals by sub-sampled data F_3/2 filters (with coefficients 1-2-1) in two directions. Note that each of the filters has a total weight of 1 + 2 + 1 = 4, and thus, it is possible to simply arithmetically shift 2 to the right instead of normalizing the division to simplify the implementation. Since each filtering unit also samples its individual input signals by a factor of 2, the signals at the individual outputs of the C 6 and D 6 filter units are actually sub-sampled by a factor of 32 in each direction ( Or reduce the frequency bandwidth by 1 0 2 4 times) 〇-19- 1353166 The next two filter units C7 and D7 are specific filter units, denoted by Fz_5. The Z suffix represents a filter that eliminates any zeros from the total normalization weighting. The general FZ_n filter equation is as follows:
Sa jj * Wjj 輸出値=_ ; (15)Sa jj * Wjj output 値 = _ ; (15)
Iwjj * 6(ajj) 其中^是2D輸入値;Wij是2D濾波器係數;及5(aij)是 函數定義如下: S(aij)=l,如果 aij#〇,否則 δ(3υ) = 0 ; (16) 如自方程式(1 5 )可見,不同於一般濾波器在於總加權不 再恰爲已知規格化常數(known normalization constant)。因 爲具有零値之進來像素(incoming pixel)的數量沒有預先知 悉’所以加權之持續累加器必需保持。同時,濾波器迴路 正在作業中,如果所對應輸入値不是零値,則累加器之內 容以現有濾波器係數之値來增加。 因爲總加權沒有固定及預先知悉,所以最後濾波器輸出 之規格化視總加權之値而定。然而,使用具有可能總加權 値之多重選擇的預定乘法表,仍然可避免方程式(15)之除 算作業。 使用特定Fz_5濾波器之目的,是當濾波器得到太接近於 半色調區之邊緣時,獲得可靠之頻率及半色調總加權預估 値。 兩個Μ X _ 5模組C 8及D 8搜索在5 X 5視窗內之最大値, 而且輸出最大値。 各該兩個插置模組C9及D9以32之因數來插置(即,向 1353166 上抽樣)信號將其回復到原析像度。各插置單元實施雙線性 插置(bilinear interpolation),尤其產生用於各4個原像素 之32*32像素。雙線性插置之步進大小是原像素光柵之32 分之一。 半色調加權模組D10接收以FRQ及FRQ_B來表示之兩 個插置單元的輸出做爲輸入。半色調加權模組將來自各輸 入之貢獻値如下述地加在一起: (17)Iwjj * 6(ajj) where ^ is 2D input 値; Wij is 2D filter coefficient; and 5(aij) is a function defined as follows: S(aij)=l, if aij#〇, otherwise δ(3υ) = 0; (16) As can be seen from equation (15), unlike general filters, the total weighting is no longer just a known normalization constant. Since the number of incoming pixels with zero 没有 is not known in advance, the weighted continuous accumulator must be maintained. At the same time, the filter loop is working. If the corresponding input 値 is not zero, the contents of the accumulator are increased by the existing filter coefficients. Since the total weight is not fixed and known in advance, the normalization of the final filter output depends on the total weighting. However, using a predetermined multiplication table with multiple choices of possible total weights 仍然, the division of equation (15) can still be avoided. The purpose of using a particular Fz_5 filter is to obtain a reliable frequency and halftone total weighted estimate when the filter gets too close to the edge of the halftone region. The two Μ X _ 5 modules C 8 and D 8 search for the maximum 値 in the 5 X 5 window, and the maximum output is 値. Each of the two interposer modules C9 and D9 is interpolated (i.e., sampled up to 1353166) by a factor of 32 to return it to the original resolution. Each interleave unit implements a bilinear interpolation, in particular 32*32 pixels for each of the four original pixels. The step size of the bilinear interpolation is one-third of the original pixel grating. The halftone weighting module D10 receives the outputs of the two interleave units represented by FRQ and FRQ_B as inputs. The halftone weighting module adds the contributions from each input as follows: (17)
HT W = HTWh + HTWl ; 其中 HTWh = (FRQ-Th)*SFh 如果 FRQ>TH ;否則 〇 ; (18) HTWL = (FRQ_B_TL)*SFL 如果 FRQ_B>TL ;否貝IJ 〇 ; (19) 其中Τη及Tl是兩個預疋臨界値’而SFh及SFl分別是原 (局)及濾波(低)頻率FRQ及FRQ_B2兩個預定比例因數。 多加邏輯確保所限制HTW値絕不超過所允許[〇,·.. 2 5 5 ] 之8-位元範圍。 第6圖說明方程式(17)、(18)、(19)及限制HTW値在所 允許範圍之多加邏輯限幅效應(clipping effect)的圖示。LA 所表不區域表不行-項目區域。如在第6圖所示,一個特別 衫色據網圖型可自所不H F Η T之位置改變到μ ρ η T到L F Η T ’因爲其頻率自高改變到中及低。因爲2D圖繪上軌跡所示 曲線是凸出形’所以單獨地按照F R Q或r q β不可能區 別濾網頻率。 參照第2圖’像素控制模組Β7接收來自濾波器Β2之超 模糊信號(super blur signal)BLR_A、來自據波器Β4之對比 1353166 値CLO、及來自半色調預估模組B6之頻率FRQ及半色調 加權HTW做爲輸入。CLO、FRQ及HTW全部是8 -位元數 ,僅超模糊信號BLR_A是全彩(L、a、b)信號。 像素控制模組B 7以一個像素一個像素方式來產生關於 濾波器陣列B 1中濾波輸出之那一對要以混合模組2 〇 6及混 合比例來混合的瞬間決定。決定經由控制信號B N K來傳送 到混合模組206來執行。在其一實施例中,BNK輸出是8-位兀信號,其最闻有效位兀(most significant bit)表示選擇 濾波器組2 0 2中之那一輸出用於混合,而其餘5個最小有 效位元(least significant bit)表示施加於所選擇濾波輸出 及連續一個(更大一號之爐波器)的混合比例。根據所期望 混合精確度來選擇有效分數位元之數。實際混合作業在混 合模組206內側使用全彩線性插置來實施。 此外,像素控制模組B7就像素中性度(pixel neutrality) 及邊緣銳化(以信號SHRP來表示),也產生多加強化控制。 8-位元信號NTL及SHRP也傳送到混合模組206來執行。 第7圖是在其一實施例所實施控制模組中所包括像素控 制模組的方塊圖示。 , 像素控制模組B 7使用三個可程式規畫之片段線性函數 來控制其作業。這些函數包括SHRP(FRQ)、KIL(CLO)及 BNK(FRQ)。通常,片段線性函數映像輸入之8-位元到輸出 之8-位元,而且可以使用完整256·輸入檢查表(full 256-entry lookup table) 來實施 。然而 ,因 爲這些 函數之 形狀非 常平滑,所以這些函數可以使用片段線性分段之小數量來 -22- 1353166 等效地近似。在其一實施例中,這些函數可以實施如同片 段線性分段,而其內容(用於實驗目的及用於調整對特別掃 描器之系統)可在周圍移動端點來調整。在另一實施例中, 爲簡化,函數使用類屬檢查表(generic lookup table)來實施。 如自第7圖可見,8-位元輸出信號SHRP以使得8-位元 輸入信號FRQ通過片段線性函數方塊702來產生。FRQ信 號也使用做爲片段線性函數方塊706之輸入,來產生預設 8-位元信號bnk。8-位元CL0輸入以片段線性函數方塊704 來映像到8 -位元信號S m ο 〇 t h。 中間bnk信號由乘算單元70 8以2之因數來相乘。然後 ,線性插置單元使用來使得來自單元7 0 8之加倍bnk輸出 和原信號bnk混合。混合量由片段線性函數方塊704所產 生之控制信號Smooth來決定。單元708及710之目的,在 當時比高時來減小濾波器大小(bnk値所示);而當對比低 (smooth)時,增加濾波器大小達到加倍。然後,8-位元混合 輸出在乘算單元712中和8-位元輸入信號HTW相乘,所得 到信號規格化來適用在8-位元範圍內,而形成輸出BNK。 如第6圖所示,低HTW意即行項目區。另一方面,高HTW 値表示可能是半色調。 中性邏輯方塊7 1 4接收超模糊信號B LR_ A及半色調加權 信號HTW。在第7圖之中性邏輯方塊714的函數如下所述 。首先,超模糊信號BLR_ A之色度平方値加在一起來形成 信號CS Q : CSQ = ( aBLR_A)2 + (bBLR_A)2 ' (20) 1353166 CSQ信號先比較臨界値TCSq,來決定其是否具有大色度 分量·‘ 若(CSQ> TCSq),貝ij NTL = 0 ; (21) 若(CSQS TCSq),則信號具有低色度,而實施後續測試: 如果[(CSQ*SFCSq+HTW)<Tntl],則 NTL=1 ;否則 NTL=0 ; (22) 其中SFcsq是預定比例因數,而TNTL_是預定常數參數。在 方程式(2)後之合理性是已知像素之中性度也受半色調加 權HTW所影響。對小HTW値(在第6圖之臨界値TH及TL 上低強度),在現有像素將表明非中性及相之亦然前,可忍 受色度平方CSQ之較大量。 參照第2圖,在其一實施例中,根據自像素控制模組B 7 所接收控制信號,混合模組2 0 6選擇來自濾波器陣列B 1 之兩個連續濾波器輸出,而且比例地使得兩者混合來產生 去除濾選輸出。 如上所述,在其一實施例中,8-位元BNK信號之格式 (format)使得8 _位元BNK信號中三個最高有效位元選擇正 確濾波器輸出,而其餘5個最小有效位元使用來決定所要 施加到本選擇濾波器輸出及連續(較大一個大小)濾波器輸 出之混合量。 第8圖是在其一實施例8 00所實施混合模組之方塊圖示 。實施例8 0 0包含:全彩線性插置單元8 1 0,來混合兩個 所選濾波器組輸出在一起;非銳光罩濾波器8 2 0,以使得 影像銳化來進一步強化混合輸出;及色度控制單元8 3 〇, 控制去除濾選輸出信號之中性。 -24- 1353166 線性插置單元8 1 0接收來自濾波器陣列B 1之5個全彩輸 出即BLR_3至BLR_11及原全彩源輸入SRC做爲輸入。在 任何已知時間,僅這些輸入中之兩個使用來混合,但是所 使用特定對可根據來自像素控制模組B7之BNK輸入來瞬 間地交換。注意6個輸入之堆疊允許混合模組2 0 6在整個 全濾波範圍’即自來濾波源C S Q到濾波器陣列B 1之最低 通濾波信號,來產生平滑變動的輸出。 混合方程式是: 輸出値=BLR_n*a + BLR_(n+l)*(l-a); (23) 其中η是BNK之三個最高有效位元之値;而a是以BNK 之其餘5個最小有效位元所表示之混合分數値(biending fraction value)。注意n = 0定義來選擇原信號SRC做爲BLR_〇 。插置是以三個彩色分量獨立地在(L、a、b)上實施之全彩 插置。 然後,來自線性插置單元8 1 0之混合輸出通過非銳遮罩 濾波器8 2 0。非銳遮罩濾波器8 2 0以使得影像銳化來進一 步強化混合輸出。銳化量由像素控制模組B 7所輸出之8 -位元SHRP信號來控制。 非銳遮罩濾波器8 2 0之作業是經由加法器8 2 2,自混合 輸出來減去原輸入之低頻版(例如,超模糊信號BLR_A)而 獲得。然後,差經由乘算單元8 2 4以S H R P信號所決定一 些因數來比例,然後經由加法器8 2 6加回到混合輸出。 因爲非銳遮罩濾波器8 2 0減去一部份低頻內容,所以差 (即,加法器826之輸出)包含較多高頻內容。使得較多高 1353166 頻內容(β卩8 24之輸出)加回到原輸出,淨結果將強化影像 而且使得其銳化。 此外’混合模組8 0 0包含色度控制單元8 3 0,其以直接 地控制色度分量(a、b)來提供控制去除濾選輸出信號之中 性的選擇。當像素決定爲中性時,輸出DSC色度分量可以 設定輸出色度値在a = b=l 28 (對應原點之色度値1 28)來強制 爲零。同樣地’如果像素決定爲非中性,輸出D S C色度分 量可以設定輸出色度値在a = b=129(對應+1之色度値129) 來強制離開零。用於中性之控制自像素控制模組B 7經由 NTL信號來傳送,如前所述。 (五)圖式簡單說明 第1圖是本發明之系統方塊圖; 第2圖是本發明系統之其一實施例; 第3圖是說明其一實施例之濾波器組中各種濾波器的一 維濾波響應; 第4圖是其一實施例之控制模組所包括半色調預估模組 的範例結構; 第5圖表示第4圖半色調預估模組內所包括最小-最大檢 測模組使用的最小-最大檢測方式; 第6圖說明在半色調預估模組中所包括半色調加權模組 實施之方程式; 第7圖是其一實施例所實施控制模組中所包括像素控制 模組的方塊圖;及 第8圖是其一實施例所實施控制模組之方塊圖示。 -26- 1353166 主要部分之代表符號說明 100,200 系 統 10 1 輸 入 信 號 102,202 濾 波 器 組 10 3 濾 波 輸 出 信 號 1 04,2 04 控 制 模 組 1 05 控 制 信 號 106,206 混 合 模 組 107,109 控 制 信 號 111 去 除 濾 CBE 之 信 號 206 混 合 模 組 708,712 乘 算 單 元 7 14 中 性 邏 輯 丨品 塊 710,810 線 性 插 置 單 元 820 非 銳 遮 罩 濾 波 器 -27HT W = HTWh + HTWl ; where HTWh = (FRQ-Th) * SFh if FRQ > TH; otherwise 〇; (18) HTWL = (FRQ_B_TL) * SFL if FRQ_B >TL; No Bay IJ 〇; (19) where Τη And Tl is two pre-critical thresholds 而 and SFh and SFl are two predetermined scaling factors of the original (bureau) and the filtered (low) frequencies FRQ and FRQ_B2, respectively. Multiple logic ensures that the limit HTW does not exceed the 8-bit range allowed [〇,·.. 2 5 5 ]. Figure 6 illustrates a graphical representation of the equations (17), (18), (19), and the limits of the HTC 値 in the allowed range plus the logic clipping effect. LA does not indicate the area table - project area. As shown in Fig. 6, a special shirt color network pattern can be changed from the position of H F Η T to μ ρ η T to L F Η T ' because its frequency changes from high to medium and low. Since the curve shown in the 2D plot is a convex shape, it is impossible to distinguish the screen frequency by F R Q or r q β alone. Referring to FIG. 2, the pixel control module Β7 receives the super blur signal BLR_A from the filter Β2, the contrast 1353166 値CLO from the Β4, and the frequency FRQ from the halftone estimation module B6. Halftone weighted HTW is used as input. CLO, FRQ and HTW are all 8-bit numbers, and only the super-fuzzy signal BLR_A is a full-color (L, a, b) signal. The pixel control module B 7 generates an instantaneous determination of the pair of filtered outputs in the filter array B 1 to be mixed by the mixing module 2 〇 6 and the mixing ratio in a pixel by pixel manner. The decision is transmitted to the hybrid module 206 via the control signal B N K for execution. In one embodiment, the BNK output is an 8-bit chirp signal, the most significant bit of which indicates that the output of the selected filter bank 2 0 2 is used for mixing, while the remaining 5 are least valid. The least significant bit indicates the mixing ratio applied to the selected filtered output and one continuous (larger number one furnace). The number of effective fractional bits is selected based on the desired blending accuracy. The actual mixing operation is performed inside the hybrid module 206 using full color linear interpolation. In addition, the pixel control module B7 also generates multi-enhanced control in terms of pixel neutrality and edge sharpening (represented by the signal SHRP). The 8-bit signals NTL and SHRP are also transmitted to the hybrid module 206 for execution. Figure 7 is a block diagram of a pixel control module included in a control module implemented in an embodiment thereof. The pixel control module B 7 uses three programmable linear functions of the program to control its operations. These functions include SHRP (FRQ), KIL (CLO), and BNK (FRQ). Typically, the fragment linear function maps the 8-bit input to the 8-bit output and can be implemented using the full 256-entry lookup table. However, because the shapes of these functions are very smooth, these functions can be similarly approximated using a small number of linear segments of the segment -22- 1353166. In one embodiment, these functions can be implemented as segment linear segments, and their content (for experimental purposes and for adjusting the system to a particular scanner) can be adjusted by moving the endpoints around. In another embodiment, for simplicity, the function is implemented using a generic lookup table. As can be seen from Figure 7, the 8-bit output signal SHRP is generated such that the 8-bit input signal FRQ is passed through the segment linear function block 702. The FRQ signal is also used as an input to the Fragment Linear Function Block 706 to generate a preset 8-bit signal bnk. The 8-bit CL0 input is mapped to the 8-bit signal S m ο 〇 t h in a segment linear function block 704. The intermediate bnk signal is multiplied by a multiplication unit 70 8 by a factor of two. The linear interpolation unit is then used to mix the doubled bnk output from unit 708 with the original signal bnk. The amount of mixing is determined by the control signal Smooth generated by the segment linear function block 704. The purpose of units 708 and 710 is to reduce the filter size (bnk値) at the time of the high ratio; and to increase the filter size to double when the contrast is low. Then, the 8-bit mixed output is multiplied in the multiplying unit 712 by the 8-bit input signal HTW, and the resulting signal normalization is applied in the 8-bit range to form the output BNK. As shown in Figure 6, low HTW means the project area. On the other hand, a high HTW 値 indicates that it may be halftone. Neutral logic block 7 1 4 receives super-blurred signal B LR_ A and halftone weighted signal HTW. The function of the neutral logic block 714 in Fig. 7 is as follows. First, the chrominance squared 超 of the super-fuzzy signal BLR_ A is added together to form the signal CS Q : CSQ = ( aBLR_A) 2 + (bBLR_A) 2 ' (20) 1353166 The CSQ signal first compares the threshold 値 TCSq to determine whether it has Large chrominance component ·' If (CSQ> TCSq), Bay ij NTL = 0; (21) If (CSQS TCSq), the signal has low chrominance and subsequent tests are performed: If [(CSQ*SFCSq+HTW)< ;Tntl], then NTL=1; otherwise NTL=0; (22) where SFcsq is a predetermined scaling factor and TNTL_ is a predetermined constant parameter. The rationality after equation (2) is that the neutrality of the known pixel is also affected by the halftone weighted HTW. For small HTW 値 (lower intensity on the critical 値TH and TL in Figure 6), the larger the amount of chrominance squared CSQ can be tolerated before the existing pixels will indicate non-neutral and phase. Referring to FIG. 2, in an embodiment, based on the control signal received from the pixel control module B7, the hybrid module 206 selects two continuous filter outputs from the filter array B1, and proportionally The two are mixed to produce a filtered output. As described above, in one embodiment, the format of the 8-bit BNK signal causes the three most significant bits of the 8-bit BNK signal to select the correct filter output, while the remaining five least significant bits Use to determine the amount of mixing to be applied to the output of the selection filter and the output of the continuous (larger size) filter. Figure 8 is a block diagram of a hybrid module implemented in an embodiment 800. Embodiment 800 includes: a full color linear interpolating unit 810 to mix two selected filter banks to output together; a non-sharp mask filter 820 to sharpen the image to further enhance the mixed output And the chromaticity control unit 8 3 〇, control removes the neutrality of the filtered output signal. The -24- 1353166 linear interpolating unit 8 1 0 receives the five full-color outputs from the filter array B 1 , namely BLR_3 to BLR_11 and the original full color source input SRC as inputs. At any known time, only two of these inputs are used for mixing, but the particular pair used can be exchanged instantaneously based on the BNK input from pixel control module B7. Note that the stacking of six inputs allows the hybrid module 206 to produce a smoothly varying output over the entire full filtering range', the self-filtering source CsQ to the lowest pass filtered signal of the filter array B1. The mixing equation is: Output 値 = BLR_n * a + BLR_(n + l) * (la); (23) where η is the three most significant bits of BNK; and a is the least significant of the remaining 5 BNK The biending fraction value represented by the bit. Note that n = 0 is defined to select the original signal SRC as BLR_〇. The insertion is a full color insertion performed independently on (L, a, b) with three color components. Then, the mixed output from the linear interpolation unit 810 passes through the non-sharp mask filter 820. The non-sharp mask filter 820 further sharpens the blended output by sharpening the image. The amount of sharpening is controlled by the 8-bit SHRP signal output by the pixel control module B7. The operation of the non-sharp mask filter 802 is obtained by subtracting the low frequency version of the original input (e.g., the super-blur signal BLR_A) from the mixed output via the adder 8 2 2 . Then, the difference is proportional to the factors determined by the S H R P signal via the multiplying unit 8 24 and then added back to the mixed output via the adder 8 2 6 . Since the non-sharp mask filter 820 subtracts a portion of the low frequency content, the difference (i.e., the output of the adder 826) contains more high frequency content. This adds more high-level 1353166 frequency content (the output of β卩8 24) back to the original output, and the net result will enhance the image and sharpen it. In addition, the hybrid module 800 includes a chrominance control unit 830 that provides a choice to control the removal of the neutrality of the filtered output signal by directly controlling the chrominance components (a, b). When the pixel is determined to be neutral, the output DSC chrominance component can be set to output chrominance 强制 at a = b = l 28 (corresponding to the chromaticity of the origin 値 1 28) to force zero. Similarly, if the pixel is determined to be non-neutral, the output D S C chrominance component can be set to output chromaticity 値 at a = b = 129 (corresponding to +1 chromaticity 値 129) to force off zero. The control for neutral is transmitted from the pixel control module B 7 via the NTL signal as previously described. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of the system of the present invention; FIG. 2 is an embodiment of the system of the present invention; FIG. 3 is a diagram illustrating various filters in a filter bank of an embodiment thereof. Dimensional filter response; FIG. 4 is an example structure of a halftone estimation module included in the control module of one embodiment; FIG. 5 is a diagram showing a minimum-maximum detection module included in the halftone estimation module of FIG. The minimum-maximum detection mode used; Figure 6 illustrates the equations implemented by the halftone weighting module in the halftone estimation module; Figure 7 is the pixel control mode included in the control module implemented in one embodiment. A block diagram of a group; and FIG. 8 is a block diagram of a control module implemented in one embodiment. -26- 1353166 Main part representation symbol 100,200 System 10 1 Input signal 102, 202 Filter bank 10 3 Filtered output signal 1 04,2 04 Control module 1 05 Control signal 106, 206 Hybrid module 107, 109 Control signal 111 Remove signal from filter CBE 206 hybrid module 708, 712 multiplication unit 7 14 neutral logic component block 710, 810 linear interpolation unit 820 non-sharp mask filter -27