Method of reconstructing image from compression-processed image signals

• US 4,807,029 A
• Filed: 06/16/1987
• Issued: 02/21/1989
• Est. Priority Date: 06/16/1986
• Status: Expired due to Term

• Current Assignee: Fujifilm Corporation (Fujifilm Holdings Corporation)
• Original Assignee: Fuji Photo Film Co Limited (Fujifilm Holdings Corporation)
• Inventors: Tanaka, Nobuyuki
• Primary Examiner(s): Groody, James J.
• Assistant Examiner(s): Peng, John K.
• Application Number:
  US07/062,799
• Time in Patent Office: 616 Days
• Field of Search: 358/133, 358/135, 358/136, 382/43, 382/56, 375/122
• US Class Current: 348/400.1
• CPC Class Codes: H04N 19/60 using transform coding

Abstract

In a method of reconstructing an image from compression-processed image signals, the image is reconstructed from transformed image signals obtained by carrying out irreversible compression encoding processing of image signals and then carrying out decoding processing. The irreversible compression encoding processing and the decoding processing of training image signals representing a predetermined training image are carried out to obtain transformed training image signals prior to the image reconstruction. The transformed training image signals are compared with the training image signals, and differences therebetween are calculated at respective picture elements. Noise signals having a signal value distribution in accordance with the distribution of the differences are generated, and the image is reconstructed by randomly adding the noise signals to the transformed image signals.

Claims

1. A method of reconstructing an image from compression-processed image signals, in which an image signal presented for reconstruction is comprised of transform image signals obtained by carrying out irreversible compression encoding processing of input image signals, and then carrying out decoding processing.

wherein the improvement comprises the steps of:

(i) carrying out said irreversible compression encoding processing and said decoding processing on input training image signals representing a predetermined training image, to obtain transformed training image signals prior to image reconstruction, comparing said transformed training image signals with said input training image signals, and calculating differences between said transformed training image signals and said input training image signals at respective picture elements,

(ii) generating noise signals having a signal value distribution in accordance with the distribution of said differences, and

(iii) reconstructing said input image by randomly adding said noise signals to said transform image signals.

2. A method as defined in claim 1 wherein an image of the same type as said image which is to be reconstructed is used as said training image.

3. A method as defined in claim 1 wherein said irreversible compression encoding processing is encoding by orthogonal transformation.

4. A method as defined in claim 1, further comprising the step of reproducing said reconstructed input image.

5. A method as defined in claim 1, wherein said irreversible compression encoding processing comprises the steps of orthogonally transforming an input image signal according to a Hadamard transform, and performing zonal sampling processing on said orthogonally transformed signal, to discard components above a predetermined high frequency.

6. A method as defined in claim 1, further comprising the step of selecting said training image from a plurality of stored training images in accordance with the attributes of the input image.

7. A method as defined in claim 1, wherein said calculating step comprises the step of determining the number of quantization levels by which picture elements of said input training image signals are smaller or larger than that of said transformed training image signal.