Patent Application
20060008167
The present invention relates to digital image capture, and more particularly to a digital camera having a picture quality setting.
Digital cameras such as the Kodak DC 280 apply JPEG compression prior to storing captured images in a camera memory card. JPEG compression involves performing a discrete cosine transform (DCT) on blocks of pixels (e.g. 8×8) of the image. The DCT coefficients are quantized to compress the image, and the quantized coefficients are entropy encoded (e.g. Huffman encoding) to produce the compressed image file. Such cameras may include a picture quality setting feature that allows the operator to select a quality option, for the image that is stored in the camera. The quality selection chooses a quantization table scale factor or quantization table used to quantize the DCT coefficients.
Generally a smaller image file results from a lower quality setting, and vice versa. One problem with this approach is that for a given quality setting images having different amounts of image detail will have different apparent quality. For example a very busy image compressed at an intermediate quality setting will have a low quality appearance, whereas an image with very little detail may have a high quality appearance even if it is compressed at the lowest quality setting.
There is a need therefore for an improved technique for accurately adjusting image quality in a digital camera.
The need is met according to the present invention by providing a method of storing an image in a digital camera, comprising the steps of: capturing an image; selecting a quantization table based on an image quality metric; and compressing the image using the selected quantization table.
Memory cards (i.e., CompactFlash card) are widely used by digital capture devices such as digital cameras to store captured images before they are transferred to other storage mediums. Due to the limited capacity of a memory card, digital cameras such as the Kodak DC 280 apply JPEG compression prior to storing captured images in a camera memory card. JPEG compression involves performing a discrete cosine transform (DCT) on blocks of pixels (e.g. 8×8) of the image. Image compression is achieved when the DCT coefficients are quantized with a quantization table. The quantized coefficients are entropy encoded (e.g. Huffman encoding) to produce the compressed image file.
Different levels of compression can be achieved with different quantization tables, and in general, the more an image is compressed, the lower the quality the image has, and vice versa. Therefore, digital cameras such as the Kodak DC 280 normally include a picture quality setting feature that allows the operator to select a quality option, essentially trading off between image quality and the number of images that can be stored in a memory card. Within the camera design, the quality selection actually chooses an appropriate quantization table used to quantize the DCT coefficients.
The current invention adds intelligence to the quantization table selection process by adding an image analysis step using certain image quality metric(s). With this image-dependent approach, consistent apparent quality may be achieved for images having different amounts of details.
Referring first to
For camera settings (i.e., quality, date) as well as for image preview, a LCD 22 is also included in a typical digital camera along with buttons 242628 for navigating within the LCD. For example, underneath the Quality Setting, a camera may let users select from three choices (as shown in
“Best”, “Better” and “Good”, respectively. A user may press up arrow button 24 or down arrow button 28 to switch among these three settings, and then press the selection button 26 for final quality selection. All these actions are coordinated by the microprocessor controller, and the quality selection by a user will eventually be feedback to the microprocessor as well.
Referred to
The compressed image is then decompressed 48 to reconstruct the processed digital image. Image quality metric(s) is(are) further applied 50 to evaluate the quality of the reconstructed processed digital image. After that, a decision has to be made whether the quality of the reconstructed processed digital image meets the user requirement 52. If the image quality indicated by the image quality metric for the specific reconstructed processed digital image is appropriate for the user selected quality settings, then the copy of the compressed image in RAM is sent to the memory card for storage 54. Otherwise, a more appropriate quantization table is selected, the copy of the processed digital image is then retrieved from RAM and it goes through compression 46, decompression 48, and image quality evaluation 50 steps again, followed with another decision step 52. Recursive loops of step 44, 46, 48, 50 and 52 might be necessary until a satisfying result can be achieved.
For a three-level quality setting of “good”, “better” and “best” with three quantization tables, namely “coarse”, “medium” and “fine”, respectively, the number of recursive loops is limited to 2, incurring additional but reasonable computation. If the initially selected quantization table produces a compressed image with consistent quality to the user selected quality setting, the compressed image is sent to the memory card for storage and no further processing is needed; if the initially selected quantization table produces a compressed image with lower quality than the user selected quality setting, a finer quantization table, if still available, is used to compress the processed digital image, until a satisfying result is achieved; if the initially selected quantization table produces a compressed image with higher quality than the user selected quality setting, a coarser quantization table, if still available, is used to compress the processed digital image, until a satisfying result is achieved. Alternatively, all the available quantization tables can be used to compress the processed digital image, and a satisfying result is selected according to the image quality metric; however this procedure is computationally inefficient compared to the recursive procedure.
Major image quality issues associated with JPEG compression are the severity of blocking and contouring artifacts in a JPEG compressed image. Therefore, any image quality metric that correlates with visual perception of blocking and contouring artifacts may be used in the image quality evaluation step. In the embodiment of the current invention, the metric used is the one described in the commonly assigned U.S. patent application Ser. No. ______ (Docket No. 81593), filed on even date herewith in the names of Q. Yu and J. Luo and entitled “A Method of Detecting the Extent of Blocking and Contouring Artifacts in a Digital Image”, which is incorporated herein by reference. This metric measures both the amount of blocking and contouring artifacts within a JPEG compressed image, and predicts the image quality that will be perceived by consumers. More specifically, this metric is based on a digital image processing method that includes the steps of: forming a column difference image; averaging the values in the columns in the column difference image to produce a column difference array; computing the average of the values in the column difference array that are separated by one block width to produce a block averaged column difference array; locating the peak value in the block averaged column difference array; calculating the mean value of the block averaged column difference array excluding the peak value to produce a column base value; computing the ratio between the peak value and the base value to produce a column ratio; repeating steps the above steps in the row direction to produce a row ratio; and employing the column and row ratios as a measure of the extent of blocking artifacts in the digital image. In an additional series of steps the extent of contouring artifacts is determined by the steps of: locating block boundaries based on the locations of peak values of column and row difference arrays; calculating a DC value for each block; generating a histogram of the block DC values; calculating the Fourier transform of the histogram; locating the first non-DC peak in the Fourier transform domain; calculating a DC quantization step size based on the frequency of the first non-DC peak; and employing the DC quantization step size as a measure of the extent of the contouring artifacts in the digital image.
Note that the entropy encoding process in the compression step and the entropy decoding process in the decompression step cancel each other and they do not affect the image quality of the reconstructed processed digital image. Therefore, for implementation efficiency, these two steps are not included in the embodiment of current invention.
The invention has been described in detail with particular reference to certain preferred embodiments thereof, but it will be understood that variations and modifications can be effected within the spirit and scope of the invention.
