Lossy Compression
Lossy compression means you will lose information and data post-compression. Furthermore, lossy compression can give you a much larger compression ratio compared to lossless compression. This may be necessary for scenarios where you must squeeze image files into a certain size for transfer. It is important to note that lossy compression schemes can give artefacts in the image when parameters for controlling compression are incorrect.
In the world of medical imaging, lossy compression is not acceptable for certain aspects, especially if the image will not provide vital diagnostic information. PACS (picture archiving and communication systems) is geared towards fixing this issue; compressing images to a perfect ratio to not lose any information while drastically changing file size. However, this form of communication is more relevant in the next page, regarding lossless compression.
Lossy image compression can take the form of various specific schemes, shown below with corresponding examples:
In the world of medical imaging, lossy compression is not acceptable for certain aspects, especially if the image will not provide vital diagnostic information. PACS (picture archiving and communication systems) is geared towards fixing this issue; compressing images to a perfect ratio to not lose any information while drastically changing file size. However, this form of communication is more relevant in the next page, regarding lossless compression.
Lossy image compression can take the form of various specific schemes, shown below with corresponding examples:
Transform-based Compression: Wavelet Transform (WT)
In transform-based compression, knowledge regarding what the image will be applied to is known. Because of this, information can be compressed in a certain manner to not affect the overall application of the image. This can be very useful when wanting to lower the size of an image considerably.
In the specific example of wavelet transform, pixels are transformed into specific coefficients. This information regarding coefficients is then compressed much more easily, because on a statistical basis, there are few groupings of different coefficients. These coefficients are now quantized and encoded to a file of smaller size. Below is an example of wavelet transform:
In DWT (discrete wavelet transform), wavelet coefficients store all the information needed to reconstruct the image. The unnecessary information is then removed, creating a satisfactory image.
In transform-based compression, knowledge regarding what the image will be applied to is known. Because of this, information can be compressed in a certain manner to not affect the overall application of the image. This can be very useful when wanting to lower the size of an image considerably.
In the specific example of wavelet transform, pixels are transformed into specific coefficients. This information regarding coefficients is then compressed much more easily, because on a statistical basis, there are few groupings of different coefficients. These coefficients are now quantized and encoded to a file of smaller size. Below is an example of wavelet transform:
In DWT (discrete wavelet transform), wavelet coefficients store all the information needed to reconstruct the image. The unnecessary information is then removed, creating a satisfactory image.
Graph showing DWT with different compression methods (right). These are reconstructions normalized using mean square error through different compression ratios.
The DWT points refer to an image that is transformed and only coefficients larger than a threshold are kept when reconstructing, while the rest of wavelet coefficients are
discarded. The DWT bands_cluster points refer to a compressed image with select c-means in wavelet reconstruction using the average compression ratio of 0.35 to 0.65. The DWT texture_cluster points refer to a compressed image with wavelet coefficients applied to
regions that are not as important.
The DWT points refer to an image that is transformed and only coefficients larger than a threshold are kept when reconstructing, while the rest of wavelet coefficients are
discarded. The DWT bands_cluster points refer to a compressed image with select c-means in wavelet reconstruction using the average compression ratio of 0.35 to 0.65. The DWT texture_cluster points refer to a compressed image with wavelet coefficients applied to
regions that are not as important.
Common JPEG Lossy Compression
JPEF compression uses discrete cosine transform to reduce image size. This converts each form of spatial domain into the frequency domain. This information is then quantized, and a compressed image can be reduced like below. Note that the Q values (quality factors) represent the parameter to tune the quality of the JPEG image (from 0-100). A lower number produces an image with maximum compression (smallest file size) but worst quality. The optimal compression Q value would vary based on the photo and application.
JPEF compression uses discrete cosine transform to reduce image size. This converts each form of spatial domain into the frequency domain. This information is then quantized, and a compressed image can be reduced like below. Note that the Q values (quality factors) represent the parameter to tune the quality of the JPEG image (from 0-100). A lower number produces an image with maximum compression (smallest file size) but worst quality. The optimal compression Q value would vary based on the photo and application.