System and method for enhancing videos from drift-free scalable bitstream

1. A scalable decoder system having a decoder reconstruction system, wherein the decoder reconstruction system comprises:

an estimation algorithm for estimating a lost discrete cosine transform (DCT) coefficient, wherein the estimation algorithm includes a density distribution dependent on at least one unknown parameter {circumflex over (α)} l ; and

a parameter estimation algorithm for estimating {circumflex over (α)} l according to the equation:

α

^

l

=

-

log

⁡

(

1

-

P

^

l

)

b

l

,

wherein b l denotes each of a set of quantizer decision levels l, and {circumflex over (P)} l is the sum of a set of a normalized frequency of occurrences of data quantized to each quantizer decision level 1,2, . . . , l.

2. The scalable decoder of claim 1 , wherein {circumflex over (P)} l is chosen to be approximately equal to one.

3. The scalable decoder of claim 1 , wherein the estimation algorithm is given by:

x

~

i

,

n

e

=

a

i

,

n

+

E

Z

⁢

{

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

}

,

and wherein Z is a quantization noise, E z is an expectation with respect to quantization noise,

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

denotes a reconstruction offset,

x

~

i

,

n

-

1

e

is an expectation of a DCT coefficient of a previous frame, ρ equals a predetermined value between −1 and 1, and a i,n is a beginning of a quantization interval.

4. The scalable decoder of claim 3 , wherein ρ equals approximately 1 for low frequency DCT coefficients, and less than 1 for high frequency DCT coefficients.

5. A parameter estimation and optimal reconstruction (PEOR) method for use in a scalable decoder, the method comprising:

determining the set of quantization occurrences;

estimating a parameter α, wherein α is estimated according to the equation

α

^

l

=

-

log

⁡

(

1

-

P

^

l

)

b

l

,

wherein b l denotes each of a set of quantizer decision levels l, and {circumflex over (P)} l is the sum of a set of a normalized frequency of occurrences of data quantized to each quantizer decision level 1,2, . . . ,l;

generating a preliminary reconstruction point according to the equation

x

~

i

,

n

e

=

a

i

,

n

+

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

)

;

and

smoothing the preliminary reconstruction point to generate an optimal reconstruction point.

6. The method of claim 5 , wherein the set of quantization occurrences are determined by counting DCT coefficients with a quantizer index of j and then dividing the result by the total number of DCT coefficients.

7. The method of claim 5 , wherein the step of smoothing the preliminary reconstruction point is accomplished according to the equation:

x

~

i

,

n

e

=

a

i

,

n

+

E

z

⁢

{

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

}

,

wherein Z is a quantization noise, E z is an expectation with respect to the quantization noise Z,

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

denotes a reconstruction offset,

x

~

i

,

n

-

1

e

is an expectation of a DCT coefficient of a previous frame, ρ equals a predetermined value between −1 and 1, and a i,n is a beginning of a quantization interval.

8. A parameter estimation and optimal reconstruction (PEOR) system for use in a scalable decoder, comprising:

a reconstruction system for determining an optimal reconstruction according to the equation:

x

~

i

,

n

e

=

a

i

,

n

+

E

z

⁢

{

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

}

,

wherein Z is a quantization noise, E z is an expectation with respect to quantization noise Z,

d

i

,

n

⁡

(

ρ

;

x

~

i

,

n

-

1

e

+

Z

)

denotes a reconstruction offset,

x

~

i

,

n

-

1

e

is an expectation of a discrete cosine transform (DCT) coefficient of a previous frame, ρ equals a predetermined value between −1 and 1, and a i,n is a beginning of a quantization interval.

9. The PEOR system of claim 8 , wherein the parameter wherein α is estimated according to the equation:

α

^

l

=

-

log

⁡

(

1

-

P

^

l

)

b

l

,

wherein b l denotes each of a set of quantizer decision levels l, and {circumflex over (P)} l is the sum of a set of a normalized frequency of occurrences of data quantized to each quantizer decision level 1,2, . . . , l.