topic 9 - image compression Flashcards
what is non - lossy (lossless) compression?
Non-lossy (lossless) compression – original image can be recovered exactly from compressed image file.
E.g. Portable Network Graphics (.png).
what is lossy compression?
Lossy compression – Information lost when file is compressed. Therefore only an approximation to the original image is possible.
Eg. Joint Experts Photographic Group JPEG (.jpg).
what is a coding redundancy?
Caused by sub-optimal code words for symbol encoding - a symbol typically (but not exclusively) represents grey-level.
what is an inter - pixel redundancy?
Inter-pixel redundancy: Due to grey-level correlations between neighbouring pixels.
what is a psycho - visual redundancy?
Information contained within an image that is superfluous to the interpretation or aesthetics of an image.
what does clarity depend on with pyscho visual redundancy?
Clarity depends on:
1) spatial frequency.
2) amplitude.
what is a symbol?
A Symbol is a general term that can refer to pixel values or transformation coefficient values.
what is the 1st component of the image coder?
The first component of the image coder is an image transformation (aka mapper) that converts the input image A(x, y) into a format that is better suited to encoding. This transforms the pixels into coefficients.
what is the 2nd component of an image coder?
The quantizer achieves compression by approximating the transformation coefficients. Quantization is an inherently lossy process whereas the image transform and coding processes are not.
what is the 3rd component of an image coder?
Symbol coding compresses the image by exploiting the fact that in natural images some grey-levels occur more frequently than others (also applies to transformation coefficients).
what is the JPEG image compression “recipe”?
- Split into 8x8 blocks and treat each block separately.
- Apply Discrete Cosine Transformation (DCT) to each image block.
- Quantise the DCT coefficients (transformed 8x8 pixel block) .
- Apply Huffman coding scheme (symbol coder) to quantised coefficients.
what is Huffman encoding?
Huffman encoding is a lossless coding method that aims to reduce the average word length of the symbols for a specified digital image. Here each symbol is initially encoded using 3 binary bits – we want to calculate a more compact representation.
what is the step by step method of Huffman encoding?
- Determine the image histogram of symbol values.
- Order the symbols by increasing probability of occurrence.
- Combine the two symbols with lowest probability.
- The above two steps are repeated until only the most probable symbol of the original image and the combined symbol remain.
- Assign the code word 1 to the most probable symbol in the image and assign 0 to the combined symbol.
