Image encoding/decoding method and device for performing feature quantization/ de-quantization, and recording medium for storing bitstream
An image encoding/decoding method and apparatus are provided. An image decoding method may comprise obtaining, from a bitstream, a feature set extracted from an input image using an artificial neural network-based feature extraction method, first information on importance of each of a plurality of channels included in the feature set and second information on a quantization method of the feature set, decoding the feature set, the first information and the second information, and dequantizing the decoded feature set based on the decoded first information and the decoded second information. The second information may comprise the number of quantization bits of each of the plurality of channels, and the number of quantization bits may be determined based on the importance of each of the plurality of channels.
1. An image decoding method performed by an image decoding apparatus, the image decoding method comprising:
obtaining, from a bitstream, a feature set extracted from an input image using an artificial neural network-based feature extraction method, first information on importance of each of a plurality of channels included in the feature set and second information on a quantization method of the feature set;
decoding the feature set, the first information and the second information; and
dequantizing the decoded feature set based on the decoded first information and the decoded second information,
wherein the second information comprises a number of quantization bits of each of the plurality of channels,
wherein the number of quantization bits is determined based on the importance of each of the plurality of channels,
wherein the first information and the second information have a decoding unit,
wherein the decoding unit is determined to be one of a sequence unit, a group of feature set (GOF) unit, a feature set unit or a channel unit, and
wherein the decoding unit is determined based on data similarity between the feature set and a previous feature set preceding the feature set in decoding order.
2. The image decoding method of claim 1 , wherein the number of quantization bits has one value selected based on the importance of each of the plurality of channels from among a plurality of preset values.
3. The image decoding method of claim 1 , wherein the second information comprises a minimum value of each of the plurality of channels, based on an activation function used to extract the feature set being a first activation function.
4. The image decoding method of claim 1 , wherein a minimum value of each of the plurality of channels is set to a lower limit value of a second activation function, based on an activation function used to extract the feature set being the second activation function.
5. The image decoding method of claim 1 , wherein the feature set is dequantized based on a predetermined quantization group comprising a plurality of quantization intervals.
6. An image decoding apparatus comprising a memory and at least one processor,
wherein the at least one processor is configured to:
obtain, from a bitstream, a feature set extracted from an input image using an artificial neural network-based feature extraction method, first information on importance of each of a plurality of channels included in the feature set and second information on a quantization method of the feature set;
decode the feature set, the first information and the second information; and
dequantize the decoded feature set based on the decoded first information and the decoded second information,
wherein the second information comprises a number of quantization bits of each of the plurality of channels,
wherein the number of quantization bits is determined based on the importance of each of the plurality of channels,
wherein the first information and the second information have a decoding unit,
wherein the decoding unit is determined to be one of a sequence unit, a group of feature set (GOF) unit, a feature set unit or a channel unit, and
wherein the decoding unit is determined based on data similarity between the feature set and a previous feature set preceding the feature set in decoding order.
7. An image encoding method performed by an image encoding apparatus, the image encoding method comprising:
extracting a feature set comprising a plurality of channels from an input image using an artificial neural network-based feature extraction method;
determining a quantization method of the feature set, based on importance of each of the plurality of channels;
quantizing the feature set based on the determined quantization method; and
encoding the quantized feature set, first information on the importance and second information on the quantization method,
wherein the second information comprises a number of quantization bits of each of the plurality of channels,
wherein the number of quantization bits is determined based on the importance of each of the plurality of channels,
wherein the first information and the second information have an encoding unit,
wherein the encoding unit is one of a sequence unit, a group of feature set (GOF) unit, a feature set unit or a channel unit, and
wherein the encoding unit is determined based on data similarity between the feature set and a previous feature set preceding the feature set in encoding order.
8. The image encoding method of claim 7 ,
wherein the first information is encoded as a difference from information on importance of a previous feature set preceding the feature set in encoding order, and
wherein the second information is encoded as a difference from information on a quantization method of the previous feature set.
9. The image encoding method of claim 7 , wherein the feature set is quantized based on a predetermined quantization group comprising a plurality of quantization intervals.
10. The image encoding method of claim 9 , wherein the quantizing the feature set comprises:
mapping each of the plurality of channels to any one of the plurality of quantization intervals;
normalizing each of the plurality of channels; and
quantizing each of the plurality of normalized channels using a minimum value and maximum value of the quantization interval mapped to each of the plurality of channels.
11. The image encoding method of claim 10 , wherein the second information comprises index information specifying the quantization interval mapped to each of the plurality of channels.
12. A computer-readable recording medium storing a bitstream generated according to the image encoding method of claim 7 .