×

Method and apparatus for reduction of image data compression noise

  • US 5,629,778 A
  • Filed: 05/15/1995
  • Issued: 05/13/1997
  • Est. Priority Date: 05/15/1995
  • Status: Expired due to Term
First Claim
Patent Images

1. A method for reducing the effects of blocking artifacts attributable to quantization incurred in the transform image coding and compression of a two-dimensional image signal into a digital image, the degree of compression determined by a scaling factor κ

  • and a quantization table, and the image signal provided as a series of electrical signals, each electrical signal corresponding to a characteristic of an element of a two-dimensional image, where the image elements have been configured as a two-dimensional H×

    V array, said method comprising the steps of;

    converting the series of electrical signals into a set of numerical values, each said numerical value quantitatively describing a feature characteristic of a corresponding image element, said numerical values further denoted by s0 (z,y) where 0≦

    y≦

    H-1 and 0≦

    z≦

    V-1;

    formatting said set of numerical values into a plurality of N×

    N image data matrices identified by indices `p` and `q`, each said image-data matrix comprising image-data terms denoted by sq,p (j,i), each said image data term comprised of a said numerical value determined in accordance with the relationship;

    
    
    space="preserve" listing-type="equation">s.sub.q,p (j,i)=s.sub.0 (j+Nq,i+Np)where 0≦

    i,j≦

    N-1;

    transforming each said image-data matrix into an N×

    N frequency-coefficient matrix comprising frequency-coefficient terms denoted by Sq,p (υ



    ), said step of transforming each said image-data matrix accomplished by means of an orthogonal transform basis matrix C in accordance with the expression,
    
    
    space="preserve" listing-type="equation">S.sub.q,p (υ



    )=C×

    s.sub.q,p (j,i)×

    C.sup.T ;

    dividing each said frequency-coefficient term Sq,p (υ



    ) by a scaled quantization term denoted by Q(υ



    ), said scaled quantization term obtained from the quantization table and modified by the scaling factor, to yield a plurality of N×

    N quotient-coefficient matrices comprising quotient terms denoted by Qu q,p (υ



    ) derived in accordance with the expression, ##EQU16## rounding each said quotient term to a less precise value to yield a quantized-coefficient matrix comprising quantized-quotient terms denoted by Qcq,p (υ



    ) in accordance with the expression, ##EQU17## subtracting each said quantized-coefficient term Qcq,p (υ



    ) from a corresponding said quotient term Quq,p (υ



    ) to form a plurality of N×

    N difference-coefficient matrices comprising difference terms Dcq,p (υ



    ) derived in accordance with the expression, ##EQU18## deriving a quantization error matrix, comprising terms E0



    ), by summing and averaging error terms, said error terms comprising functions of said difference-coefficient matrices;

    selecting a set of filter parameters, denoted by α and

    β

    ;

    multiplying each said quantized coefficient term by a corresponding quantization term to yield a plurality of N×

    N mask-multiplied transform coefficient matrices comprising terms denoted by Rq,p (υ



    ) derived in accordance with the expression,
    
    
    space="preserve" listing-type="equation">R.sub.q,p (υ



    )=Qc.sub.q,p (υ





    Q(υ



    );

    applying an inverse orthogonal transform to said mask-multiplied transform coefficient matrices to yield a plurality of N×

    N received image-data matrices denoted by rq,p (j,i), in accordance with the inverse transform equation,
    
    
    space="preserve" listing-type="equation">r.sub.q,p (j,i)=C.sup.T ×

    R.sub.q,p (υ





    C;

    formatting said received image-data matrices into an H×

    V received image-data set comprising terms denoted by sR (z,y), in accordance with the equation,
    
    
    space="preserve" listing-type="equation">s.sub.R (j+Nq,i+Np)=r.sub.q,p (j,i)where 0≦

    i+Np≦

    H-1 and 0≦

    j+Nq≦

    V-1;

    forming said received image data set into a plurality of N×

    N overlapped image-data matrices comprising terms denoted by vs,r (j,i), in accordance with the equation,
    
    
    space="preserve" listing-type="equation">v.sub.s,r (j,i)=s.sub.R (j+ω

    s, i+ω

    r);

    transforming said overlapped image-data matrices into modified coefficient matrices comprising terms denoted by Svs,r (υ



    ), in accordance with the equation,
    
    
    space="preserve" listing-type="equation">Sv.sub.s,r (υ



    )=C×

    v.sub.s,r (j,i)×

    C.sup.T ;

    converting said modified coefficient matrices into filtered coefficient matrices by means of said quantization error matrix and said filter parameters, said modified coefficient matrix comprising terms denoted by Sfs,r (υ



    ), said step of converting said modified coefficient matrices performed in accordance with the equation, ##EQU19## transforming said filtered coefficient matrices into filtered image-data matrices in accordance with the transform equation,
    
    
    space="preserve" listing-type="equation">rf.sub.s,r (j,i)=C.sup.T ×

    Sf.sub.s,r (υ





    C; and

    converting said filtered image-data matrices into a series of filtered electrical signals such that said filtered electrical signals can be configured into a two-dimensional H×

    V array of filtered image elements, each said filtered electrical signal corresponding to a characteristic of one said filtered image element.

View all claims
  • 13 Assignments
Timeline View
Assignment View
    ×
    ×