200829027 九、發明說明： 【發明所屬之技術領域】 本發明係有關於-種壓縮裝置，聽尤指影縮裝置及其方法。 【先前技術】 按’隨著電·迅速魏财聽麟的絲，許錄位化資訊 ，顧I„age)、聲音(Audl〇)、視訊(Vide〇)及文件細胸在網路迅 、便捷的傳輸，並且以數位多媒體⑼伽她㈣ =腦的儲存裝置中’藉以達到更穩定，有效且長久的資料保存。通常多 ^體應用的資料量非常域大，因此多媒體龍必須經過壓縮才可縮短傳 ，間，其賴内視鏡也是必須經種縮才可縮短傳輸時間，但又不能使 衫像失真。 在膠囊内視鏡的應用中’考量電池壽命與效能間的取捨是—侧艮非常 重要的因素。目前最先進的視訊壓縮技術能夠勤本雜高的壓縮率來降 低影像資料龍日杨位元率’由於這_演算法需要大量的計算量來達成 =壓縮率的效能’但是卻會造成大量的功率損耗，連帶影響電池的壽命。 *此’需多快速的壓縮演算法紛紛被發展出來降低運算時的計曾量，但3 卻會造成壓縮過後«的影像畫面與原始未壓縮的影:: 明顯的失真造成。這樣《後影《面非料適合用於醫療診斷， 也係M RGB三原色為基礎的影像驗演算法可以使記憶體與計算量的需求 大幅降低’同時，可以使減少膠囊内視鏡中射頻(RF)端的功^耗，= 處理器本身之壓鮮可以制最小獅轉4:1的條件限制。、“ 請參閱第-圖，其係為習知技術之影像壓縮震置之方塊圖，如圖所示， 其包含―影像感應器（Image sensor)10，一第一壓縮單元a， 〜曰 、'、百早兀14，、一第三壓縮單元16，以及一第四壓縮單元18,。其設計B 接針對影像感應器捕捉的原始200829027 IX. DESCRIPTION OF THE INVENTION: TECHNICAL FIELD OF THE INVENTION The present invention relates to a compression device, particularly a shadow reduction device and method thereof. [Prior Art] Press 'With the electricity, quickly Wei Cai listens to Lin's silk, Xu recorded information, Gu I„age), sound (Audl〇), video (Vide〇) and file thin chest on the Internet, Convenient transmission, and digital multimedia (9) gamma (four) = brain storage device 'to achieve more stable, effective and long-term data preservation. Usually multi-body application data volume is very large, so multimedia dragon must be compressed It can shorten the transmission, and the endoscope must also be shortened to shorten the transmission time, but it can not make the shirt image distortion. In the application of the capsule endoscope, the consideration between the battery life and the performance is - side艮 very important factor. At present, the most advanced video compression technology can reduce the image data, Long Riyang bit rate, because of the high compression ratio. Because this algorithm requires a lot of calculation to achieve the performance of compression ratio, but It will cause a lot of power loss, which will affect the battery life. * This 'faster compression algorithm has been developed to reduce the amount of calculations, but 3 will cause compression after the imagery Face and original uncompressed shadow:: Obvious distortion caused. This kind of "back image" is not suitable for medical diagnosis, but also the M RGB three primary color based image verification algorithm can make the memory and calculation demand greatly Lowering 'at the same time, can reduce the power consumption of the radio frequency (RF) end of the capsule endoscope, = the compression of the processor itself can make the minimum lion turn 4:1 condition limit., "Please refer to the figure - the system The block diagram of the image compression shock of the prior art, as shown in the figure, includes an "Image sensor" 10, a first compression unit a, ~曰, ', 百早兀14, 一一A third compression unit 16, and a fourth compression unit 18,. Its design B is connected to the original captured by the image sensor
Bayer影像值執行像素分離的動作產生疋 貝料、第-綠色貧料、第二綠色資料以及藍色資料四個子未處理之影像資 200829027 料（Raw image)畫面分量後，針對這四個子晝面分量各別由第一壓縮單元 12’ 、第二壓縮單元14，、第三壓縮單元16,以及第四壓縮單元丨8，分別 進行麵。其各別由離散餘弦轉換單元完成8對8 _之二轉散餘二轉 換（Discrete Cosine TranSform)l2〇，、140’ 、160’ 、180，、再經量化 單元量化（Quantlzat1〇n)122，、142，、162，、182，、編碼單元12工4里、 144 、164 、184進行熵編碼(Entr〇py c〇ding)等壓縮過程的運算，其 中熵編碼採用非查表與較低計算複雜度的Le_el—Ziv⑺以呢。 惟查’目前的影像壓_算法_對於512χ512 均為256x256x8位凡，而二維離散餘弦轉換也同樣為8χ8矩陣 片«跚G· 18um聰製程在h 8V的操作電壓下二 =功率為R 9Μ，射_物肖耗3•麵料 ====_=新纖壓縮裝置咖法，不 了改。目樣㈣視鏡消耗辨依然偏高，可解決上述之問題。 【發明内容】 $明之主要目的，在於提供_種景彡 單 衫像貢料進行次取樣，以達到低消耗功率之目的。_由取樣早兀對 -第三取樣單元。其壓縮方法係:一:樣:兀、-第二壓縮單元，以及 色資料，並Μ縮紅色資料，產^物早70 ’接收一影像資料之-紅 像資料之_第_綠色資料 輸出資料’第-取樣單元，接收影 -綠色資料，產生—第二輪出資料弟=綠色資料，第二_單元，壓縮第 二綠色資料，並取樣第二綠色2取樣單元，接收影像資料之-第 、λ弟—壓縮單元，壓縮第二綠色資料， 200829027 產生一第三輸出資料，第三取樣單元，接收影像資料之一藍色資料，並取 樣藍色資料，產生一第四輪出資料。 【實施方式】 茲為使貴審查委員對本發明之結構特徵及所達成之功效有更進一步 之瞭％與遇識，謹佐以較佳之實施例及配合詳細之說明，說明如後： 本發明構想係以第一圖之影像壓縮演算法為架構，分析影像二維離散 餘弦轉換的能量分佈及DC/AC係數的變異性，根據分析的結果，我們發現 在腸胃道的影像中紅色資料之分量的平均直流係數最大，其次是第一綠色 資料與第二綠色資料之分量，最後是藍色資料之分量。在交流係數的平均 變異數分析部分，紅色資料之分量無論是在水平、垂直和對角線均有最大 的能量’其次S第-綠色資料與第二綠色資料之分量，最後是藍色資料之 分量。因此，我們將影像次取樣的技巧加人第—綠色資料、第二綠色資料 ，藍色資料之分量，減少影像壓縮處理器第—綠色資料、第二綠色資料與 監色=料之分量的記憶體需求量，降低影像處理器消耗功率。 =㈣第_圖，其係為本發明之—較佳實施例之方塊圖。如圖所示， 3 == 像感應器1(3、—第―壓縮單元12、—第—取樣單元14、 弟一Μ备目單兀16、一楚一百▽採⑽-ΊΟ » 弟—取樣早兀18、一第三壓縮單 資料四個子未處理之f像資=Γ讀弟’、、表色貪料以及一藍色 接收影像_认色=(rRaw image)晝面分量彳_—壓縮單元12， 取樣單元，接收亚壓縮紅说料，產生—第―輪出資料，第- 料，第二壓縮單元/61 %巴貧料’並以2:1次取«一綠色貿 2:1次取樣第二綠色資料，第 -第二輸出資料，第二::Γ取樣單元14取樣後之第1色資料，產生 ，-…一 取樣早元18，接收影像資料之第二綠色資料，並以 7 200829027 之藍色資料，並以4:1次取樣藍色資料，產生一第四輸出資料。 其中，第一壓縮單元12更包括一離散餘弦轉換單元120、一掃瞄單元 122、一量化單元124以及一編碼器126。離散餘弦轉換單元12〇，轉換影 像資料之紅色資料，也就是離散餘弦轉換單元120要讓空間上的資料轉換 到頻率值域，因為在頻率值域之資料通常會集中在低頻，而且人眼本來就 對高頻的訊號較不敏感，因此可以藉由離散餘弦轉換單元120以減少空間 贅餘量。 掃瞄單元122，掃描離散餘弦轉換單元120所轉換之紅色資料，產生一 第一掃瞄資料，其中在本實施例中係使用鋸齒形排列（zig—Zag scan)方式 進行掃描，其就是將一個區塊(block)資料依低頻至高頻排列，這是因為通 常數值都是集中在低頻附近，而之後編碼單元必須要該些數值及中才 能有效壓縮。 量化單元124，量化掃瞄單元122所產生之接收第一掃瞄資料，產生一 第一量化資料，此量化單元的目的即是控制壓縮縮率與失真的主要關鍵， 編碼器126，編碼第一量化資料，產生第一輸出資料，其中編碼器126係為 熵編碼器’以編碼第一量化資料。 同理’第二壓縮單元16與第三壓縮單元2〇係包括一離散餘弦轉換單 兀160，200、一掃瞄單元162，2〇2、一量化單元164，2〇4以及一編碼器 166 ’ 206。其中離散餘弦轉換單元丨6〇，2〇〇係以4對4距陣對第一綠色資 料與第二綠色資料進行離散餘弦轉換，其上述之離散餘弦轉換單元16〇， 200、掃瞒單元162，202、量化單元164，204以及編碼器166，206作用與 作動關係與上述相同，故此不再多加贊述。 &承上所述’本發明是在影像壓縮處理裝置的第二壓縮單元16與第三壓 縮單疋20前中分別加入第一取樣單元14與第二取樣單元以2:1次取樣技 \由於貪料量因為第_取樣單元14與第二取樣單元18對第—綠色資料 與第二綠色資料做次取樣而減少，因此將第一圖之習知技術的第二壓縮單 凡14與第三壓縮單元16’之離散餘弦轉換單元14〇，，16〇，的8χ8二維 8 200829027 =弦轉換改為本發明之離散餘弦轉換單元⑽，的4χ4二轉散餘 22州魏樣_碰，便不做任何_ °Λ_ 視鏡，_單元12之雜餘弦轉 壓縮έ ^衫析紅色貞料的結果對於整各影像的減相當4要，所以此 η彳Γ料之步驟與第―圖之習知技術相同。最後因為藍色資料直接傳 輸π有任何_縮’在次取樣技術大幅度的減少輸人的龍量。又 Ζ壓縮率，因此在第—_單元12、第二壓縮單元16與第三壓縮單元 20加入掃描單元122、162、2〇2以增加壓縮率。 接上所述，本實施例之次取樣方法可由數學公式表示，其如下所示： BMxeami1 mod 4，j mod 4) m = 1，2,3,4,5,6,7,8· ^Mw2m ^ u{m-\) u(m-5) u(m-2) u(m-6) U(m — T) u(m - 3) u(m - 8) u(m - 4) u(m-2) u(m-5) w(m-l) u(m-6) _u{m-l) u(m-3) w(w-8) u(m-4) 其中為對次取樣率為16-to-2m的次取樣遮罩，u(n)係為步階函數 (Step Function)。 本發明之一較佳貫施例由於加入第一取樣單元1心第二取樣單元與 第二取樣單元22以次取樣技術大幅度的減少第一綠色資料、第二綠色資料 及藍色資料之輸入資料量，進而大大的降低了記憶體及邏輯所消耗之功 率。實際的晶片使用TSMC 0.18um 1P6M製程在1，8V的操作電壓下 PrimePower™估計之總消耗功率為9· 17mW，較目前膠囊内視鏡的影像壓縮處 理裝置降低38.53%，而平均？81^1從原來的32.51(^略微下降至32.18(^， 對於醫生來說影像平均之PSNR大於30dB已可接受。壓縮率則從79· 65%降 至79· 62% ’依然符合75%的標準。 9 200829027 睛一併參閲第三圖，係為本發明之一較佳實施 首先由影像感應器ίο捕捉的原始影像資料執行像素分離1的^ :圖所示， 資料、-第一綠色資料、一第二綠色資料以及_ 生一紅色 影像資料㈣卿e編分量後，執行步_;—^鮮未處理之 像貧料之紅色資料，並壓縮紅色資料，產生第一輪出資料早12接收衫 色資料中，更包括離散餘弦轉換影像資料之紅色二^、中於«縮紅 後之紅色資料，產生m倾，量化第_掃^散餘弦轉換 化資料，編碼第-量化資料，產生第一輸出資料。、’、，產生一第一量 再執行步驟S12接收影像資料之一第一綠色 料，其中係由第-取樣單元14以2:1次取樣第!: t ^取樣第一綠色資 ⑽縮取樣後之第一綠色資料，產生一第二:出 驟S10相同，故不再多加贊述。再接:私方式與步 色資料，並卿：綠纖，其幅====一第二綠 3色資料，接下來執行步驟S18 _取樣後之第二綠色:1 -輪出貧料，其中壓縮之方式與步驟sl〇 、產生弟 來，行步謂接收影像資料之一藍色資料述=訂 四輪峨’其中係由第三取樣單元22以4:1次^^^ 〜社舰’本發明之影賴難置包含—第_ 、 早兀、—第二壓縮單元、-第二取樣單 ^ 一第一取樣 取樣單元。其_方法係由第-_=：接:^^元，以及一第三 並壓縮紅_，產生—第—輸峨 ^貝^-紅色資料， —第-綠色資料，並取樣第-綠色資料，第接收影像資料之 輸出資料，第:取樣單元，魏影ς料 科，絲樣第二綠色資料，第三_單元 $-、、·亲色貝 三輪出資料，第三取樣單元，接收影像資貧料，產生一第 以達到低雜轉之目的. l像讀進行次取樣’ 200829027 本發明係實為-具有新賴性、進步性及可 之=要件無疑，法提出發_申請應 明二凡ί it發明之—較佳實施例而已，並非用來限定本發 月貝細之辄圍’舉凡依本發明申請專利範圍所述之 神所為之均等變化與修飾，均應包括於本發明之f請專利範_ 【圖式簡單說明】 第一圖係為習知技術之影像壓縮裝置之方塊圖； 第二圖係林發明之-健實_之麵圖丨以及 第三圖係為本發明之一較佳實施例之流程圖。 【主要元件符號說明】 10， 影像感應器 12， 第一壓縮單元 120, 離散餘弦轉換單元 122, 量化單元 124， 編碼單元 14， 第二壓縮單元 140， 離散餘弦轉換單元 142， 量化單元 144， 編碼單元 16, 第三壓縮單元 160， 離散餘弦轉換單元 162， 量化單元 164， 編碼單元 18， 第四壓縮單元 200829027 180， 離散餘弦轉換單元 182， 量化單元 184， 編碼早元 10 影像感應器 12 第一壓縮單元 120 離散餘弦轉換單元 122 掃描單元 124 量化單元 126 編碼單元 14 第一取樣單元 16 第二壓縮單元 160 離散餘弦轉換單元 162 掃描單元 164 量化單元 166 編碼早元 18 第二取樣單元 20 第三壓縮單元 200 離散餘弦轉換單元 202 掃描單元 204 量化單元 206 編碼單元 22 第三取樣單元The Bayer image value performs the pixel separation operation to generate the 疋 料 , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , The components are respectively surfaced by the first compression unit 12', the second compression unit 14, the third compression unit 16, and the fourth compression unit 丨8. The discrete cosine transform unit performs a Discrete Cosine TranSform l2〇, 140', 160', 180, and a quantized unit quantization (Quantlzat1〇n) 122, respectively. , 142, 162, 182, and encoding unit 12, 144, 164, 184 perform entropy coding (Entr〇py c〇ding) and other compression processes, wherein entropy coding uses non-lookup and lower calculation The complexity of Le_el-Ziv (7). However, the current image pressure_algorithm_ is 256x256x8 for 512χ512, and the 8D8 cosine transform is also 8χ8 matrix. «跚G· 18um Cong process at h 8V operating voltage 2 = power is R 9Μ , shooting _ material Xiao consumption 3 • fabric ====_= new fiber compression device coffee method, can not be changed. The objective (4) the consumption of the mirror is still high, which can solve the above problems. SUMMARY OF THE INVENTION The main purpose of the invention is to provide a sub-sampling of a single-shirt tribute to achieve low power consumption. _ by sampling early - the third sampling unit. The compression method is: one: sample: 兀, - the second compression unit, and the color data, and the red data is collapsed, the product is early 70 'received one image data - the red image data _ the first _ green data output data 'The first sampling unit, receiving the shadow-green data, generating the second round of the data brother = green data, the second unit, compressing the second green data, and sampling the second green 2 sampling unit, receiving the image data - the first , λ brother - compression unit, compressing the second green data, 200829027 generates a third output data, the third sampling unit receives blue data of one of the image data, and samples the blue data to generate a fourth round of data. [Embodiment] For the purpose of making the reviewer's structural features and the effects achieved by the reviewer of the present invention further and %, the description of the preferred embodiment and the detailed description will be given as follows: Based on the image compression algorithm of the first image, the energy distribution of the two-dimensional discrete cosine transform and the variability of the DC/AC coefficient are analyzed. According to the analysis results, we find the component of the red data in the image of the gastrointestinal tract. The average DC coefficient is the largest, followed by the component of the first green data and the second green data, and finally the component of the blue data. In the analysis of the average variation of the AC coefficient, the component of the red data has the largest energy in the horizontal, vertical and diagonal lines, followed by the component of the S-green data and the second green data, and finally the blue data. Component. Therefore, we add the technique of image sub-sampling to the first-green data, the second green data, and the blue data component, and reduce the memory of the image compression processor--green data, the second green data, and the color of the material. The amount of physical demand reduces the power consumption of the image processor. = (d) Figure _ is a block diagram of a preferred embodiment of the present invention. As shown in the figure, 3 == Image sensor 1 (3, - the first - compression unit 12, - the first - sampling unit 14, the younger one, the other, the other, the first one, the other one, the other one - Sampling as early as 18, a third compression single data, four sub-unprocessed f-images = Γ reading brother',, color matching and a blue receiving image _ recognition color = (rRaw image) 彳 surface component 彳 _ _ The compression unit 12, the sampling unit, receives the sub-compressed red material, generates the first-round data, the first material, the second compression unit/61% barter material and takes 2:1 times to take a green trade 2: Sampling the second green data, the first-second output data, the second:: the first color data sampled by the sampling unit 14, generating, -... a sampling early 18, receiving the second green data of the image data, And taking the blue data of 7 200829027 and sampling the blue data 4:1 times to generate a fourth output data, wherein the first compression unit 12 further includes a discrete cosine transform unit 120, a scan unit 122, and a quantization. The unit 124 and an encoder 126. The discrete cosine transform unit 12 〇 converts the red data of the image data, that is, the discrete cosine The conversion unit 120 wants to convert the spatial data to the frequency range, because the data in the frequency range is usually concentrated in the low frequency, and the human eye is inherently less sensitive to the high frequency signal, so the discrete cosine transform unit can be used. The scan unit 122 scans the red data converted by the discrete cosine transform unit 120 to generate a first scan data, wherein in the embodiment, a zig-Zag scan is used. The method of scanning is to arrange a block of data according to low frequency to high frequency, because usually the values are concentrated near the low frequency, and then the coding unit must have these values and the medium to be effectively compressed. 124. The first scan data generated by the quantization scanning unit 122 is generated to generate a first quantized data. The purpose of the quantization unit is to control the main key of compression reduction and distortion, and the encoder 126 encodes the first quantitative data. Generating a first output data, wherein the encoder 126 is an entropy encoder to encode the first quantized data. Similarly, the second compression unit 16 and the third The compression unit 2 includes a discrete cosine transform unit 160, 200, a scan unit 162, 2, 2, a quantization unit 164, 2〇4, and an encoder 166 '206. The discrete cosine transform unit 丨6〇, 2〇〇 is a discrete cosine transform of the first green data and the second green data by using 4 pairs of 4 matrix, the discrete cosine transforming unit 16〇, 200, the broom unit 162, 202, the quantizing units 164, 204, and the above The action of the encoders 166, 206 is the same as that described above, and therefore no further comments are made. & The invention is in the front of the second compression unit 16 and the third compression unit 20 of the image compression processing apparatus. Adding the first sampling unit 14 and the second sampling unit respectively to the 2:1 sampling technique, because the first sampling unit 14 and the second sampling unit 18 subsample the first green data and the second green data. Reduction, so the second compression unit 14 of the prior art of the first figure and the discrete cosine conversion unit 14〇, 16〇 of the third compression unit 16' are changed to the invention by the 8χ8 two-dimensional 8 200829027=string conversion. Discrete cosine transform unit (10) 4χ4二转散散22州魏样_碰, do not do any _ °Λ _ mirror, _ unit 12 of the cosine to compression έ ^ shirt analysis of the red dip results for the entire image of the reduction of 4, Therefore, the steps of this data are the same as those of the prior art of the first figure. Finally, because the blue data directly transmits π, there is any _ shrinking in the sub-sampling technique to greatly reduce the amount of input. Further, the compression ratio is increased, so that the scanning unit 122, 162, 2〇2 is added to the first unit 12, the second compression unit 16, and the third compression unit 20 to increase the compression ratio. In addition, the subsampling method of this embodiment can be expressed by a mathematical formula as follows: BMxeami1 mod 4,j mod 4) m = 1,2,3,4,5,6,7,8· ^Mw2m ^ u{m-\) u(m-5) u(m-2) u(m-6) U(m - T) u(m - 3) u(m - 8) u(m - 4) u (m-2) u(m-5) w(ml) u(m-6) _u{ml) u(m-3) w(w-8) u(m-4) where is the sampling rate The 16-to-2m sub-sampling mask, u(n) is a step function. A preferred embodiment of the present invention greatly reduces the input of the first green data, the second green data, and the blue data by sub-sampling techniques by adding the second sampling unit and the second sampling unit 22 of the first sampling unit 1 The amount of data, which in turn greatly reduces the power consumed by the memory and logic. The actual chip uses the TSMC 0.18um 1P6M process at an operating voltage of 1,8V. PrimePowerTM estimates the total power consumption to be 9·17mW, which is 38.53% lower than the current image compression processing device for capsule endoscopes, and the average? 81^1 from the original 32.51 (^ slightly decreased to 32.18 (^, for the doctor, the average PSNR of the image is more than 30dB acceptable. The compression rate is reduced from 79. 65% to 79.62%' still meets 75% 9 200829027 The third figure is a preferred embodiment of the present invention. First, the original image data captured by the image sensor ίο performs pixel separation 1 : as shown in the figure, data, - first green After the data, a second green data, and _ raw red image data (4) Qing e coded, the implementation of step _; - ^ fresh untreated red material like poor material, and compressed red data, the first round of data generated early 12 receiving shirt color data, including the red cosine of the discrete cosine transform image data, the red data in the «redundant red, generating m tilt, quantizing the first _ sweep ^ cosine transform data, encoding the first - quantized data, Generating a first output data, ',, generating a first amount and then performing step S12 to receive one of the first green materials of the image data, wherein the first sampling unit 14 samples 2:1 times!: t ^ sampling The first green data after the sampling of a green capital (10) One second: the same as S10, so no more praise is added. Then: private mode and step color data, and Qing: green fiber, its width ==== a second green 3 color data, the next step S18 _Second green after sampling: 1 - Turn out the poor material, where the compression method and the step sl〇, the generation of the brother, the step is to receive one of the image data blue data description = book four rounds 其中The third sampling unit 22 includes 4:1 times ^^^ ～ the ship's shadow of the present invention includes - the first _, the early 兀, the second compression unit, the second sampling unit, the first sampling sampling unit The method is based on the -_=:: ^^ element, and a third and compressed red _, the resulting - the first - the 峨 ^ ^ ^ - red data, - the first - green data, and the sample - green Data, output data of the received image data, the first: sampling unit, Wei Ying and the Department of Materials, the second green data of the silk sample, the third _ unit $-,, · the color of the three rounds of data, the third sampling unit, receiving The image is poor, and the first one is to achieve the purpose of low miscellaneous conversion. l Read and sample the same as '200829027 The present invention is true - with new reliance and progress And the stipulations of the present invention are not limited to the preferred embodiment of the present invention, and are not intended to limit the scope of the present invention. The equal change and modification of God should be included in the invention patent _ [Simple description of the drawing] The first picture is a block diagram of the image compression device of the prior art; the second picture is the invention of the forest - The figure of the hard surface and the third figure are flowcharts of a preferred embodiment of the present invention. [Description of main component symbols] 10, image sensor 12, first compression unit 120, discrete cosine transform unit 122, Quantization unit 124, encoding unit 14, second compression unit 140, discrete cosine transform unit 142, quantization unit 144, encoding unit 16, third compression unit 160, discrete cosine transform unit 162, quantization unit 164, coding unit 18, fourth Compression unit 200829027 180, discrete cosine transform unit 182, quantization unit 184, encoding early 10 image sensor 12 first compression unit 120 discrete cosine transform Element 122 scanning unit 124 quantization unit 126 encoding unit 14 first sampling unit 16 second compression unit 160 discrete cosine transform unit 162 scanning unit 164 quantization unit 166 encoding early element 18 second sampling unit 20 third compression unit 200 discrete cosine conversion unit 202 scanning unit 204 quantization unit 206 coding unit 22 third sampling unit