JP3979020B2 - Image processing apparatus for sampling compressed image - Google Patents

Description

この（１）式において、ｍは、図２に示した８×８成分の周波数成分ブロック２１における水平方向の周波数成分（ｍ＝０〜７）を表し、ｎは、垂直方向の周波数成分（ｎ＝０〜７）を表し、ｉ、ｊは、その８×８周波数成分ブロック２１の逆ＤＣＴ演算で得られる８×８画素の画素ブロック内の個々の画素の水平方向位置（ｉ＝０〜７）と垂直方向位置（ｊ＝０〜７）を表す。また、Ｃ_ｍは、水平方向周波数成分がｍのときの変換係数を表し、Ｃ_ｎは、垂直方向周波数成分がｎのときの変換係数を表し、Ｓ_ｍｎは、８×８周波数成分ブロック２１における周波数成分（ｍ，ｎ）での周波数成分値を表す。また、Ｂは、元のビットマップ画像が持つ各色データのビット数に応じた定数を表し、Ｐ_ｉｊは、８×８周波数成分ブロック２１の逆ＤＣＴ演算で得られる８×８画素の画素ブロックにおける位置（ｉ，ｊ）での画素値を表す。
【００２１】
ＪＰＥＧ画像をプリスキャンするとき、従来は、既に述べたように、ＪＰＥＧ画像を完全に元のビットマップ画像に復元してからプリスキャンを行う。すなわち、従来は、ＪＰＥＧ画像をプリスキャンするとき、１つの８×８画素ブロックについて言えば、全６４画素の画素値を得るべく（１）式の計算を６４×６４＝４０９６回行い、その計算を１画像を構成する多数のブロックについて繰り返すので、全体の計算量は膨大である。
【００２２】
しかし、ＪＰＥＧ画像を完全に復元して１画像全ての画素値を得なくても、画像統計値は取得できるものと考えられる。別の言い方をすれば、画像統計値を取得するために、ＪＰＥＧ画像を完全に復元して膨大な数の全ての画素値を得る必要は必ずしもなく、統計計算に必要な或る程度の数の画素値が得られれば、自動画像補正等の目的に必要十分な画像統計値は取得できるものと考えられる。具体的な例として、自動画像補正のプリスキャンの場合、元の画像のサイズに関係無しに、最低６００×６００画素程度の画素の値が取得できれば良いと考えられる。例えば、１２００×１２００画素又は１８００×１２００画素のサイズを持つ画像から画像統計値を取得するためには、縦横それぞれ１つ置きに画素を取得して合計６００×６００画素又は９００×６００画素のように、上述の最低の画素数以上の数の画素の値が取得できれば良いと考えられる。
【００２３】
そこで、本実施形態では、次のようにしてＪＰＥＧ画像のプリスキャンを行う。
【００２４】
図３は、本実施形態に係るＪＰＥＧ画像のサンプリング方法の流れを示す。本実施形態に係るサンプリング方法は、ソフトウェアを実行するコンピュータ、ＡＳＩＣのような純粋なハードウェア回路、又はそれらの組み合わせなど、ソフトとハードをどのように使った構成でも実施することができる。
【００２５】
本実施形態に係るサンプリング方法では、まず、サンプリング対象のＪＰＥＧ画像に対し、ハフマン復号化（Ｓ１６）、ランレングスハフマン復号化（Ｓ１７）、及び逆量子化（Ｓ１８）を行って、図２に示したような８×８成分の周波数成分ブロック２１を、元の画像の８×８画素ブロックの各々について得る。次に、８×８画素ブロックの各々について、全６４画素のうちどの位置（ｉ，ｊ）の画素値を得るかを決定する、つまり、画素値取得対象の画素位置（ｉ，ｊ）を選択する（Ｓ１９）。そして、選択した画素位置（ｉ，ｊ）の画素のみについて、逆ＤＣＴ演算（Ｓ２０）を行い、逆ＤＣＴ演算によって得られた表色系ＹＵＶの画素値Ｐ_ｉｊを所定の表色系（例えば、ＲＧＢ又はＣＭＹ）に色変換して（Ｓ２１）、色変換して得られる画素値から画像統計値を取得する（Ｓ２２）。
【００２６】
この実施形態に係るサンプリング方法は、元画像の８×８画素ブロックの各々について、画素値取得対象の画素を８×８画素ブロックの全６４画素の中から選択し、選択した画素のみについて、逆ＤＣＴ演算及び色変換を施して画素値を得る。このため、ＪＥＰＧ画像を元画像に完全に復元してサンプリングを行う従来の方法に比べて、逆ＤＣＴ演算及び色変換の計算回数を減らすことができ、それにより、ＪＰＥＧ画像のサンプリングにかかる装置の処理負担及び処理時間長を減らすことができる。特に、逆ＤＣＴ演算の計算回数を減らすことができるのは、本効果の大きな要因である。
【００２７】
画素値取得対象の画素を選択する方法としては、種々の方法が考えられるが、以下、画素値８×８画素ブロックの各々について、８×８画素のうち所定の比率で画素を残すように間引き（以下、その比率を「間引き率」と言う）、その間引きによって残された画素を選択するという方法（以下、間引き選択方法）を例に、本実施形態のサンプリング方法及びそれの具体的な効果を説明する。なお、ここで言う「間引き率」とは、８×８画素ブロックの画素を縦横どれだけの比率（割合）で残すように画素を間引くかを表すものであり、間引き率は、画像統計値を取得するための統計計算に必要な画素数に応じて決定される。
【００２８】
例えば、画像統計値を取得するための統計計算に６００×６００画素が必要であるとき、元の画像サイズが１２００×１２００画素であれば、間引き率は、「縦横１／２（２分の１）」と決定され、また、元の画像サイズが１８００×１８００画素であれば、間引き率は、「縦横１／３（３分の１）」と決定される。
【００２９】
本実施形態のサンプリング方法は、間引き率が縦横１／２のときは、８×８画素ブロック３１の８×８画素のうち、縦横それぞれ１／２を乗じた４×４画素、計１６画素が例えば分散的に残るように、８×８画素ブロック３１の画素を間引く。具体的な例としては、図４に示すように、本実施形態のサンプリング方法は、縦方向（垂直方向）について、水平方向位置ｉ＝１、３、５、７に属する画素を残すように（つまり等間隔に画素を残すように）、他の水平方向位置ｉ＝０、２、４、６に属する画素を全て間引く。同様に、横方向（水平方向）について、垂直方向位置ｊ＝１、３、５、７に属する画素を残すように（つまり等間隔に画素を残すように）、他の垂直方向位置ｉ＝０、２、４、６に属する画素を全て間引く。
【００３０】
以上のようにして、間引き率が縦横１／２のときは、８×８画素ブロック３１の全６４画素のうち１６画素が分散的に残る。本実施形態に係るサンプリング方法は、それら１６画素のみについて画素値を算出するので、このときの逆ＤＣＴ演算及び色変換の計算回数は、従来に比べて、１６画素／６４画素（６４画素分の１６画素）、つまり、１／４（４分の１）回に減る。この記載からわかるように、間引き率を縦横１／２という比較的大きな値にしても、本実施形態の効果が十分に得られる。
【００３１】
なお、上述のことを別の観点からすれば、元のビットマップ画像を２×２画素ブロックに分割し、個々の２×２画素ブロックの中から代表的な１画素を抽出したと言える。より一般的な言い方をすれば、間引き率が「縦横１／ｋ（ｋ分の１）」のときは、元のビットマップ画像をｋ×ｋ画素ブロックに分割し、ｋ×ｋ画素ブロックの中から代表的な１画素を抽出すると言える。
【００３２】
次に、間引き率が「縦横１／８（８分の１）」とされた場合を例に説明する。
【００３３】
間引き率が「縦横１／８（８分の１）」とされた場合は、図５に示すように、本実施形態のサンプリング方法は、８×８画素ブロック３１の８×８画素のうち、縦横それぞれ１／８を乗じた１×１画素、つまり１つの画素だけ残るように、８×８画素ブロック３１の画素を間引く。具体的な例としては、図５に示すように、本実施形態のサンプリング方法は、縦方向（垂直方向）について、８列のうち或る１列に属する画素（例えば水平方向位置がｉ＝０の画素）のみを残すように他の７列に属する画素（つまり水平方向位置がｉ＝１〜７の画素）を全て間引く。同様に、横方向（水平方向）について、８列のうち或る１列に属する画素（例えば垂直方向位置がｊ＝０の画素）のみを残すように他の７列に属する画素（つまり垂直方向位置がｊ＝１〜７の画素）を全て間引く。なお、このようにｉ＝ｊ＝０の画素を残すのは単なる一例であり、他の任意の位置の画素、例えば、８×８画素ブロック３１の中央のｉ＝ｊ＝３の画素を残すようにしても良い。
【００３４】
以上のように、間引き率が縦横１／８の場合は、８×８画素ブロック３１から画素が１つだけ残される。本実施形態では、その１つの画素のみについて画素値が算出され、１画像を構成する多数の８×８画素ブロック３１における各々の１つの画素値から、画像統計値を取得するための統計計算が行われる。これについて、別の言い方をすれば、間引きによって残された１画素から得られる画素値を、８×８画素ブロック３１全体の代表的な画素値として、統計計算が行われる。この観点からすれば、８×８画素ブロック３１の代表的な画素値として、特定の位置の画素値ではなく、８×８周波数成分ブロックのＤＣ成分のみを用いて、８×８画素ブロック３１の平均的な画素値を取得することができる。そうすると、逆ＤＣＴ演算を全く行う必要なく、ＤＣ成分の演算のみで、８×８画素ブロック３１の代表的な画素値を取得することができる。
【００３５】
このＤＣ成分の演算方法について、図６を参照して具体的に説明すると、先頭の周波数成分ブロック２１ＡのＤＣ成分の値ＤＣ［Ａ］は、差分符号化された値ΔＤＣ［Ａ］に所定値０を加算することで復号し、周波数成分ブロック２１ＢのＤＣ成分の値ＤＣ［Ｂ］は、差分符号化されたΔＤＣ［Ｂ］に、上記復号した直前の周波数成分ブロック２１ＡのＤＣ［Ａ］を加算して復号し、周波数成分ブロック２１ＣのＤＣ成分の値ＤＣ［Ｃ］は、差分符号化された値ΔＤＣ［Ｃ］に、上記復号した直前の周波数成分ブロック２１ＢのＤＣ［Ｂ］を加算して復号する。
【００３６】
このようにして復号したＤＣ成分は、逆量子化前の（つまりＹＵＶ表色系で表された）、８×８画素ブロック全体の平均的な画素値を示す。従って、ＤＣ成分を復号した後は、復号したＤＣ成分に表される値に対して、逆量子化を施し、逆ＤＣＴ演算を一切行う必要なく色変換処理を行って、所定の表色系（ＲＧＢ又はＣＭＹ）で表された画素値を得れば良い。
【００３７】
以上のことから、本実施形態のサンプリング方法において間引き率を縦横１／８とした場合は、従来と比べて、逆量子化及び色変換を行うための計算回数は、１画素／６４画素（６４画素分の１画素）＝１／６４（６４分の１）回と大幅に減らすことができ、更に、最も大きな処理負担である逆ＤＣＴ演算は、一度も行う必要がない。ゆえに、本実施形態のサンプリング方法において、間引き率を縦横１／８とすれば、プリスキャンを実行する装置の負担を大幅に軽くすることができ、以って、プリスキャンに要する時間長を大幅に短縮することができるようになる。間引き率が縦横１／８となるのは、上述した説明から言えば、例えば画像統計値を取得するための統計計算に６００×６００画素が必要であるとき、元の画像サイズが縦横８倍の４８００×４８００画素であるときであるが、元の画像サイズがそれ以上であって間引き率が縦横１／８より小さくて済む場合であっても、上記効果から、８×８画素ブロックあたりの間引き率は縦横１／８として処理することができる。
【００３８】
以上、本発明の好適な実施形態を説明したが、これは本発明の説明のための例示であって、本発明の範囲をこの実施形態にのみ限定する趣旨ではない。本発明は、他の種々の形態でも実施することが可能である。
【００３９】
例えば、上述した実施形態は、ＪＰＥＧ画像に対する自動画像補正処理におけるプリスキャンとして使用されるが、自動画像補正は、元画像の全ての領域は勿論、ユーザが選択した一部の領域のみに対しても行うことができる。
【００４０】
また、上述した実施形態のサンプリング方法は、ＪＰＥＧ画像に対する自動画像補正処理におけるプリスキャンとして使用されるものであるが、本発明は、それのみに限らず、プリスキャンを必要とする種々の画像処理におけるプリスキャンとして使用できる。また、本発明のサンプリング方法は、或る画像処理の本処理前のプリスキャンとしてではなく、元画像の特定の情報（例えば画像統計値）を得ることそれ自体を目的とした単独の処理として使用することもできる。
【００４１】
また、上記実施形態では、画素を間引くときは、８×８画素ブロックに画素が分散的に残るように間引くが、必ずしも、分散的に残るように間引かなくても良い（例えば、８×８画素ブロックの或る一領域に属する画素が偏重的に残るように間引いても良い（一例として、８×８画素ブロックの右半分の領域に属する全３２画素が残るように左半分の領域に属する残り全３２画素を間引いても良い））。また、元画像を構成する個々の８×８画素ブロック或るいは元画像全体から、ランダムに、画素値取得対象の画素を選択し、選択した画素について画素値を求めるようにしても良い。
【図面の簡単な説明】
【図１】（ａ）は、ＪＰＥＧ方式による画像圧縮の概略手順を示す図。（ｂ）は、ＪＰＥＧ画像の復元の概略手順を示す図。
【図２】８×８の周波数成分ブロックを示す図。
【図３】本発明の一実施形態に係るプリスキャン方法の流れを示す図。
【図４】本発明の一実施形態において間引き率を「縦横1／2」としたときの間引きの様子を示す図。
【図５】本発明の一実施形態において間引き率を「縦横1／8」としたときの間引きの様子を示す図。
【図６】ＪＰＥＧ方式による画像圧縮において、ＤＣ成分の符号化の流れを示す図。
【符号の説明】
２１ ８×８周波数成分ブロック
３１ ８×８画素ブロック[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an improvement in an image processing apparatus that performs sampling for acquiring a predetermined type of image information of an original image from a compressed image.
[0002]
[Prior art]
In various types of image processing, before performing this processing, specific information of the image to be processed (for example, statistical values such as brightness and saturation over the entire image, hereinafter referred to as “image statistical values”) There is a thing which needs to perform sampling (hereinafter referred to as “pre-scan”) to obtain the signal. One of them is automatic image correction which is a kind of photo retouch processing. Automatic image correction optimizes image brightness, saturation, color balance, etc. in this process based on image statistics acquired by pre-scanning, such as brightness, saturation, and brightness histograms and contrast. Turn into.
[0003]
In automatic image correction, in order to obtain image statistics by pre-scanning, it is necessary to read the image to be processed. However, if the read image is a compressed image obtained by compressing a bitmap image, the original image is compressed. After completely restoring the original bitmap image, pre-scanning is performed on the original bitmap image. That is, when displaying a compressed image with automatic image correction, the compressed image is completely restored to the original bitmap image, pre-scanned to obtain image statistics, and then this processing is performed again. Start decompression of the compressed image, and during the restoration, correct the value of the restored pixel based on the image statistics obtained by prescan, and then output the corrected bitmap image The process flow is executed. That is, when the compressed image is automatically corrected, the process for completely restoring the compressed image is performed twice in combination with the pre-scan and the main process.
[0004]
[Problems to be solved by the invention]
The process for decompressing a compressed image generally requires a large amount of calculation and is a burden on the apparatus.
[0005]
For example, when the compressed image is an image compressed by the JPEG baseline method (ISO / IEC10918-1) (hereinafter referred to as a JPEG image), as is well known, DCT (discrete cosine transform) calculation is performed during image compression. When the JPEG image is restored, an inverse DCT operation is performed. The amount of calculation in the DCT operation is enormous, and the amount of calculation in the inverse DCT operation that is in an integrated relationship with it is also enormous. For this reason, the process for restoring the JPEG image is heavy on the apparatus and takes time. In addition, when a JPEG image is displayed after automatic image correction, as described above, it is necessary to restore the JPEG image twice, so that the processing load and the processing time length are doubled. This problem occurs when, for example, a large JPEG image taken with a digital camera is taken into a printer and processed by a CPU in the printer, and the compressed image is restored by a relatively low-speed CPU. , Especially large.
[0006]
Therefore, an object of the present invention is to reduce the processing load for sampling and displaying a compressed image and to shorten the processing time length thereof.
[0007]
[Means for Solving the Problems]
An image processing apparatus according to a first aspect of the present invention is an image processing apparatus that performs sampling for obtaining a predetermined type of image information of an original image from a compressed image, and is based on the total number of pixels of the original image. Pixel value calculation means for calculating pixel values from compressed image data only for a plurality of pixels that are a small number of pixels and are distributed over the entire processing target area of the original image, and pixel value calculation Image information acquisition means for acquiring the image information from pixel values of a plurality of pixels from the means.
[0008]
According to the image processing device according to the first aspect of the present invention, when acquiring predetermined types of image information (for example, statistical values such as brightness and saturation over the entire image), the number is less than the total number of pixels of the original image. For only a plurality of pixels, pixel values are calculated from the compressed image data. Thereby, compared with the conventional apparatus which calculates a pixel value about all the pixels which an original image has, the calculation amount for acquiring image information can be reduced. For this reason, the processing load for sampling and displaying the compressed image is reduced, and the processing time length thereof is shortened.
[0009]
A second image processing apparatus according to the present invention is an image processing apparatus that performs sampling for obtaining a predetermined type of image information of an original image from a compressed image, and divides the original image into k × k pixel blocks. A representative pixel is extracted from each k × k pixel block, and a pixel value calculation unit that calculates a pixel value from compressed image data only for the extracted pixel, and a representative value from the pixel value calculation unit Image information acquisition means for acquiring the image information from the pixel value of each pixel.
[0010]
According to the image processing device according to the second aspect of the present invention, when acquiring predetermined types of image information, the original image is extracted from each k × k pixel block when divided into k × k pixel blocks. For only representative pixels, pixel values are calculated from the compressed image data. Thereby, compared with the conventional apparatus which calculates a pixel value about all the pixels which an original image has, the calculation amount for acquiring image information can be reduced. For this reason, the processing load for sampling and displaying the compressed image is reduced, and the processing time length thereof is shortened.
[0011]
In a preferred embodiment according to the second aspect of the present invention, when the compressed image is a JPEG image compressed by the JPEG baseline method, the pixel value calculation means includes a frequency component block corresponding to the k × k pixel block. Among them, the pixel value of the representative pixel of the k × k pixel block is calculated using only the DC component.
[0012]
The functions of the above-described units included in the image processing apparatus according to the first and second aspects of the present invention can be implemented by a computer. A computer program therefor includes various media such as disk-type storage, semiconductor memory, and communication network. Can be installed or loaded into the computer.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the sampling method according to an embodiment of the present invention will be described by taking as an example the case where the compressed image is a JPEG image compressed by the JPEG baseline method. The sampling method of this embodiment is used as a prescan in automatic image correction processing for a JPEG image, for example.
[0014]
In this embodiment, a JPEG image is pre-scanned by a method to be described later. First, in order to make the effects produced by the pre-scan easier to understand, first, as is well known to those skilled in the art, compression (encoding) by the JPEG baseline method is performed. ) And the procedure of restoration (decoding) will be schematically described.
[0015]
FIG. 1A shows a schematic procedure for image compression by the JPEG method, and FIG. 1B shows a schematic procedure for restoration of a JPEG image.
[0016]
As shown in FIG. 1A, the general procedure for image compression by the JPEG method is to first convert the color space (usually RGB) of the bitmap image to be compressed into YUV (YC _b C _r ) (step S1) An image represented by YUV (YUV bitmap image) is converted into a pixel block of a predetermined size (for example, a square pixel block of N × N pixels, hereinafter, the pixel block is determined from the most standard 8 × 8 pixel value). (S2). Next, each pixel block is subjected to a DCT operation (S3), and a frequency component block (coefficient block) generated by the DCT operation is quantized using a different quantization step for each transform coefficient (the quantization step is a quantization step). Prepared in advance as a table) (S4). Then, the quantized transform coefficient of each coefficient block is divided into a DC coefficient and an AC coefficient, the AC coefficient is subjected to zigzag scanning, and then run-length Huffman coding is performed (S5). The HPEG encoding is performed after the differential encoding is performed (S6), and a JPEG image is generated (that is, JPEG image compression is completed).
[0017]
The general procedure for JPEG image restoration is the reverse of this general procedure for image compression.
[0018]
That is, as shown in FIG. 1B, the general procedure for restoring a JPEG image is to first perform Huffman decoding on the DC (direct current) component of the JPEG image (S7), and on the AC (alternating current) component. Run-length Huffman decoding is then performed (S8). As a result, quantized 8 × 8 frequency component blocks are obtained. Next, inverse quantization (S9) and inverse DCT calculation (S10) are performed on each frequency component block to make each frequency component block a pixel block represented by YUV, and each pixel block represented by YUV By combining (S11), a YUV bitmap image is formed. Then, the color space YUV of the image is converted into the original color space (for example, RGB) (S12), and the restoration of the JPEG image is completed.
[0019]
The above is the outline procedure for restoring a JPEG image. In this procedure, the inverse DCT calculation has the largest processing load, and the processing takes time. The following formula (1) shows the formula of inverse DCT calculation.
[0020]
[Expression 1]

In the equation (1), m represents a horizontal frequency component (m = 0 to 7) in the 8 × 8 component frequency component block 21 shown in FIG. 2, and n represents a vertical frequency component (n I, j are the horizontal positions (i = 0 to 7) of the individual pixels in the pixel block of 8 × 8 pixels obtained by the inverse DCT calculation of the 8 × 8 frequency component block 21. ) And the vertical position (j = 0 to 7). C _m represents a transform coefficient when the horizontal frequency component is m, C _n represents a transform coefficient when the vertical frequency component is n, and S _mn is the 8 × 8 frequency component block 21. The frequency component value at the frequency component (m, n) is represented. B represents a constant corresponding to the number of bits of each color data included in the original bitmap image, and P _ij is a pixel block of 8 × 8 pixels obtained by the inverse DCT operation of the 8 × 8 frequency component block 21. The pixel value at the position (i, j) is represented.
[0021]
When a JPEG image is pre-scanned, conventionally, as already described, the JPEG image is completely restored to the original bitmap image and then pre-scanning is performed. In other words, conventionally, when pre-scanning a JPEG image, for one 8 × 8 pixel block, the calculation of equation (1) is performed 64 × 64 = 4096 times to obtain a pixel value of all 64 pixels, and the calculation is performed. Is repeated for a large number of blocks constituting one image, so that the total calculation amount is enormous.
[0022]
However, it is considered that the image statistical value can be acquired without completely restoring the JPEG image and obtaining the pixel values of all the images. In other words, in order to obtain image statistics, it is not always necessary to fully restore a JPEG image to obtain a huge number of all pixel values, but to some extent necessary for statistical calculations. If pixel values are obtained, it is considered that image statistics necessary and sufficient for purposes such as automatic image correction can be acquired. As a specific example, in the case of pre-scanning for automatic image correction, it is considered that a pixel value of at least about 600 × 600 pixels may be acquired regardless of the original image size. For example, in order to acquire image statistics from an image having a size of 1200 × 1200 pixels or 1800 × 1200 pixels, every other pixel in the vertical and horizontal directions is acquired, and a total of 600 × 600 pixels or 900 × 600 pixels is obtained. In addition, it is considered that it is only necessary to obtain the value of the number of pixels equal to or more than the above-mentioned minimum number of pixels.
[0023]
Therefore, in the present embodiment, JPEG image pre-scanning is performed as follows.
[0024]
FIG. 3 shows a flow of a sampling method for JPEG images according to this embodiment. The sampling method according to the present embodiment can be implemented by any configuration using software and hardware, such as a computer executing software, a pure hardware circuit such as ASIC, or a combination thereof.
[0025]
In the sampling method according to the present embodiment, first, Huffman decoding (S16), run-length Huffman decoding (S17), and inverse quantization (S18) are performed on the JPEG image to be sampled, as shown in FIG. An 8 × 8 component frequency component block 21 is obtained for each 8 × 8 pixel block of the original image. Next, for each of the 8 × 8 pixel blocks, it is determined which position (i, j) of the total 64 pixels is to be obtained, that is, the pixel position (i, j) from which the pixel value is to be acquired is selected. (S19). Then, only for the pixels of the selected pixel position (i, j), the inverse DCT operation (S20) performed, the pixel values _{P ij} of the predetermined color system color system YUV obtained by the inverse DCT operation (e.g., (RGB or CMY) (S21), and image statistical values are acquired from the pixel values obtained by the color conversion (S22).
[0026]
In the sampling method according to this embodiment, for each 8 × 8 pixel block of the original image, a pixel value acquisition target pixel is selected from all 64 pixels of the 8 × 8 pixel block, and only the selected pixel is inverted. A pixel value is obtained by performing DCT calculation and color conversion. For this reason, the number of times of inverse DCT calculation and color conversion can be reduced compared with the conventional method in which a JEPG image is completely restored to an original image and sampling is performed. The processing burden and the processing time length can be reduced. In particular, the fact that the number of inverse DCT calculations can be reduced is a major factor of this effect.
[0027]
Various methods are conceivable as a method for selecting a pixel from which a pixel value is to be acquired. Hereinafter, for each pixel value 8 × 8 pixel block, thinning is performed so that pixels are left at a predetermined ratio of 8 × 8 pixels. (Hereinafter, the ratio is referred to as a “decimation rate”), and a method of selecting pixels left by the decimation (hereinafter, “decimation selection method”) as an example, the sampling method of this embodiment and its specific effects Will be explained. The “decimation rate” referred to here indicates how much the pixels of an 8 × 8 pixel block are to be thinned out so that the pixels are thinned out at a ratio (ratio). It is determined according to the number of pixels required for statistical calculation for acquisition.
[0028]
For example, when 600 × 600 pixels are required for statistical calculation for obtaining image statistics, if the original image size is 1200 × 1200 pixels, the thinning rate is “vertical 1/2 (1/2) If the original image size is 1800 × 1800 pixels, the thinning rate is determined as “1/3 (vertical and horizontal) (1/3)”.
[0029]
In the sampling method of this embodiment, when the decimation rate is 1/2 in the vertical and horizontal directions, among the 8 × 8 pixels in the 8 × 8 pixel block 31, 4 × 4 pixels multiplied by 1/2 in the vertical and horizontal directions, a total of 16 pixels are obtained. For example, the pixels of the 8 × 8 pixel block 31 are thinned out so as to remain in a distributed manner. As a specific example, as shown in FIG. 4, the sampling method of the present embodiment leaves pixels belonging to the horizontal position i = 1, 3, 5, 7 in the vertical direction (vertical direction) ( In other words, all the pixels belonging to other horizontal positions i = 0, 2, 4, and 6 are thinned out so as to leave the pixels at equal intervals. Similarly, in the horizontal direction (horizontal direction), other vertical position i = 0 so that the pixels belonging to the vertical position j = 1, 3, 5, 7 are left (that is, the pixels are left at equal intervals). All pixels belonging to 2, 4, 4 are thinned out.
[0030]
As described above, when the thinning rate is ½ in the vertical and horizontal directions, 16 pixels out of all 64 pixels in the 8 × 8 pixel block 31 remain in a distributed manner. Since the sampling method according to the present embodiment calculates pixel values for only those 16 pixels, the number of times of inverse DCT calculation and color conversion at this time is 16 pixels / 64 pixels (64 pixels worth) compared to the conventional case. 16 pixels), that is, 1/4 (1/4) times. As can be seen from this description, the effect of this embodiment can be obtained sufficiently even if the thinning rate is set to a relatively large value of 1/2 in the vertical and horizontal directions.
[0031]
From another point of view, it can be said that the original bitmap image is divided into 2 × 2 pixel blocks, and one representative pixel is extracted from each 2 × 2 pixel block. In more general terms, when the decimation rate is “vertical / horizontal 1 / k (1 / k)”, the original bitmap image is divided into k × k pixel blocks, It can be said that one representative pixel is extracted from
[0032]
Next, a case where the thinning rate is “vertical and horizontal 1/8 (1/8)” will be described as an example.
[0033]
When the thinning rate is “vertical 1/8 (1/8)”, as shown in FIG. 5, the sampling method of the present embodiment uses the 8 × 8 pixels in the 8 × 8 pixel block 31. The pixels of the 8 × 8 pixel block 31 are thinned out so that only 1 × 1 pixel multiplied by 1/8 in the vertical and horizontal directions, that is, only one pixel remains. As a specific example, as shown in FIG. 5, the sampling method of the present embodiment is a pixel belonging to a certain one of eight columns (for example, the horizontal position is i = 0) in the vertical direction (vertical direction). All the pixels belonging to the other seven columns (that is, the pixels whose horizontal position is i = 1 to 7) are thinned out so as to leave only the pixels. Similarly, in the horizontal direction (horizontal direction), the pixels belonging to one of the eight columns (for example, the pixel whose vertical direction position is j = 0) are left, and the pixels belonging to the other seven columns (that is, the vertical direction). All pixels whose positions are j = 1 to 7 are thinned out. It should be noted that leaving the pixel with i = j = 0 in this way is merely an example, and the pixel at any other position, for example, the pixel with i = j = 3 in the center of the 8 × 8 pixel block 31 is left. Anyway.
[0034]
As described above, when the thinning rate is 1/8 in the vertical and horizontal directions, only one pixel is left from the 8 × 8 pixel block 31. In the present embodiment, the pixel value is calculated for only that one pixel, and statistical calculation for obtaining the image statistical value from each one pixel value in the many 8 × 8 pixel blocks 31 constituting one image is performed. Done. In other words, statistical calculation is performed using the pixel value obtained from one pixel left by thinning as a representative pixel value of the entire 8 × 8 pixel block 31. From this point of view, the representative pixel value of the 8 × 8 pixel block 31 is not the pixel value at a specific position, but only the DC component of the 8 × 8 frequency component block. An average pixel value can be acquired. Then, it is not necessary to perform the inverse DCT calculation at all, and the representative pixel value of the 8 × 8 pixel block 31 can be acquired only by the calculation of the DC component.
[0035]
This DC component calculation method will be described in detail with reference to FIG. 6. The DC component value DC [A] of the first frequency component block 21A is set to a differentially encoded value ΔDC [A] by a predetermined value. The value DC [B] of the DC component of the frequency component block 21B is decoded by adding 0, and the DC [A] of the frequency component block 21A immediately before the decoding is added to the differentially encoded ΔDC [B]. The DC component value DC [C] of the frequency component block 21C is added to the differentially encoded value ΔDC [C] and the DC [B] of the frequency component block 21B immediately before the decoding is added. To decrypt.
[0036]
The DC component decoded in this manner indicates an average pixel value of the entire 8 × 8 pixel block before dequantization (that is, expressed in the YUV color system). Therefore, after the DC component is decoded, the value represented by the decoded DC component is subjected to inverse quantization, color conversion processing is performed without performing any inverse DCT operation, and a predetermined color system ( What is necessary is just to obtain the pixel value represented by RGB or CMY).
[0037]
From the above, in the sampling method of the present embodiment, when the thinning rate is set to 1/8 vertical and horizontal, the number of calculations for performing inverse quantization and color conversion is 1 pixel / 64 pixels (64 1 pixel) = 1/64 (1/64) times, and the inverse DCT calculation, which is the largest processing burden, need not be performed once. Therefore, in the sampling method of the present embodiment, if the thinning rate is 1/8 in the vertical and horizontal directions, the burden on the apparatus that performs the prescan can be greatly reduced, and the time length required for the prescan is greatly increased. Can be shortened. The decimation rate is 1/8 in the vertical and horizontal directions. From the above description, for example, when 600 × 600 pixels are required for statistical calculation for obtaining image statistical values, the original image size is 8 times vertical and horizontal. Even when the image size is 4800 × 4800 pixels, and the original image size is larger than that and the thinning-out rate can be smaller than 1/8 in the vertical and horizontal directions, the thinning out per 8 × 8 pixel block can be obtained from the above effect. The rate can be processed as 1/8 vertical and horizontal.
[0038]
The preferred embodiment of the present invention has been described above, but this is an example for explaining the present invention, and the scope of the present invention is not limited to this embodiment. The present invention can be implemented in various other forms.
[0039]
For example, the above-described embodiment is used as a pre-scan in the automatic image correction process for a JPEG image. However, the automatic image correction is performed not only on all areas of the original image but also on only a part of the areas selected by the user. Can also be done.
[0040]
In addition, the sampling method of the above-described embodiment is used as a pre-scan in automatic image correction processing for a JPEG image. However, the present invention is not limited thereto, and various image processing that requires pre-scan. Can be used as a pre-scan. Further, the sampling method of the present invention is not used as a pre-scan before a main process of a certain image process, but as a single process for the purpose of obtaining specific information (for example, image statistical values) of the original image itself. You can also
[0041]
Further, in the above embodiment, when thinning out pixels, thinning is performed so that the pixels remain in the 8 × 8 pixel block in a distributed manner, but it is not always necessary to thin out the pixels in a distributed manner (for example, 8 × 8). It may be thinned out so that pixels belonging to a certain area of the pixel block remain in an unbalanced manner (for example, it belongs to the left half area so that all 32 pixels belonging to the right half area of the 8 × 8 pixel block remain) All the remaining 32 pixels may be thinned out)). Alternatively, a pixel value acquisition target pixel may be randomly selected from individual 8 × 8 pixel blocks constituting the original image or the entire original image, and the pixel value may be obtained for the selected pixel.
[Brief description of the drawings]
FIG. 1A is a diagram showing a schematic procedure of image compression by the JPEG method. FIG. 5B is a diagram illustrating a schematic procedure for restoring a JPEG image.
FIG. 2 is a diagram showing an 8 × 8 frequency component block.
FIG. 3 is a flowchart showing a pre-scan method according to an embodiment of the present invention.
FIG. 4 is a diagram showing a thinning state when the thinning rate is “vertical and horizontal 1/2” in the embodiment of the present invention.
FIG. 5 is a diagram showing a state of thinning when the thinning rate is “vertical and horizontal 1/8” in one embodiment of the present invention.
FIG. 6 is a diagram illustrating a flow of encoding a DC component in image compression by the JPEG method.
[Explanation of symbols]
21 8 × 8 frequency component block 31 8 × 8 pixel block

Claims (3)

A prescan for acquiring image statistical information from a JPEG image compressed by a JPEG baseline method that divides an original image into k × k pixel blocks and performs DCT operation for each k × k pixel block, and the JPEG An image processing apparatus that performs an image restoration and a main process of correcting an original image obtained by the restoration based on image statistical information acquired by the prescan,
For the pre-scan,
A thinning rate determining means for determining a thinning rate based on the number of pixels of the original image and the number of pixels necessary for obtaining the image statistical information;
From each of the k × k pixel blocks, the remaining pixels are extracted by thinning out the pixels according to the thinning rate determined by the thinning rate determining unit, and the inverse DCT operation is performed only on the extracted pixels, and the JPEG image A pixel value calculating means for calculating the pixel value by restoring the data of
Image statistical information acquisition means for acquiring the image statistical information from the pixel value of the pixel calculated by the pixel value calculation means;
An image processing apparatus comprising: A prescan for acquiring image statistical information from a JPEG image compressed by a JPEG baseline method that divides an original image into k × k pixel blocks and performs DCT operation for each k × k pixel block, and the JPEG A computer program for image processing that restores an image and performs main processing for correcting the original image obtained by the restoration based on image statistical information acquired by the prescan,
For the pre-scan,
Determining a thinning rate based on the number of pixels of the original image and the number of pixels necessary to obtain the image statistical information;
From each of the k × k pixel blocks, pixels left by thinning out pixels according to the thinning rate determined by the step of determining the thinning rate are extracted, and an inverse DCT operation is performed only on the extracted pixels, Restoring JPEG image data and calculating pixel values;
Obtaining the image statistical information from the pixel value of the pixel calculated in the step of calculating the pixel value;
Computer program to run. A prescan for acquiring image statistical information from a JPEG image compressed by a JPEG baseline method that divides an original image into k × k pixel blocks and performs DCT operation for each k × k pixel block, and the JPEG An image processing method for restoring an image and performing a main process of correcting an original image obtained by the restoration based on image statistical information acquired by the prescan,
For the pre-scan,
Determining a thinning rate based on the number of pixels of the original image and the number of pixels necessary to obtain the image statistical information;
From each of the k × k pixel blocks, pixels left by thinning out pixels according to the thinning rate determined by the step of determining the thinning rate are extracted, and an inverse DCT operation is performed only on the extracted pixels, Restoring JPEG image data and calculating pixel values;
Obtaining the image statistical information from the pixel value of the pixel calculated in the step of calculating the pixel value;
An image processing method.

Images